 Manual merge from 5.0-rpl, of fixes for:
1)
BUG#25507 "multi-row insert delayed + auto increment causes
duplicate key entries on slave" (two concurrrent connections doing
multi-row INSERT DELAYED to insert into an auto_increment column,
caused replication slave to stop with "duplicate key error" (and
binlog was wrong), and BUG#26116 "If multi-row INSERT
DELAYED has errors, statement-based binlogging breaks" (the binlog
was not accounting for all rows inserted, or slave could stop).
The fix is that: in statement-based binlogging, a multi-row INSERT
DELAYED is silently converted to a non-delayed INSERT.
This is supposed to not affect many 5.1 users as in 5.1, the default
binlog format is "mixed", which does not have the bug (the bug is
only with binlog_format=STATEMENT).
We should document how the system delayed_insert thread decides of
its binlog format (which is not modified by this patch):
this decision is taken when the thread is created
and holds until it is terminated (is not affected by any later change
via SET GLOBAL BINLOG_FORMAT). It is also not affected by the binlog
format of the connection which issues INSERT DELAYED (this binlog
format does not affect how the row will be binlogged).
If one wants to change the binlog format of its server with SET
GLOBAL BINLOG_FORMAT, it should do FLUSH TABLES to be sure all
delayed_insert threads terminate and thus new threads are created,
taking into account the new format.
2)
BUG#24432
"INSERT... ON DUPLICATE KEY UPDATE skips auto_increment values".
When in an INSERT ON DUPLICATE KEY UPDATE, using
an autoincrement column, we inserted some autogenerated values and
also updated some rows, some autogenerated values were not used
(for example, even if 10 was the largest autoinc value in the table
at the start of the statement, 12 could be the first autogenerated
value inserted by the statement, instead of 11). One autogenerated
value was lost per updated row. Led to exhausting the range of the
autoincrement column faster.
Bug introduced by fix of BUG#20188; present since 5.0.24 and 5.1.12.
This bug breaks replication from a pre-5.0.24/pre-5.1.12 master.
But the present bugfix, as it makes INSERT ON DUP KEY UPDATE
behave like pre-5.0.24/pre-5.1.12, breaks replication from a
[5.0.24,5.0.34]/[5.1.12,5.1.15]
master to a fixed (5.0.36/5.1.16) slave! To warn users against this when
they upgrade their slave, as agreed with the support team, we add
code for a fixed slave to detect that it is connected to a buggy
master in a situation (INSERT ON DUP KEY UPDATE into autoinc column)
likely to break replication, in which case it cannot replicate so
stops and prints a message to the slave's error log and to SHOW SLAVE
STATUS.
For 5.0.36->[5.0.24,5.0.34] replication or 5.1.16->[5.1.12,5.1.15]
replication we cannot warn as master
does not know the slave's version (but we always recommended to users
to have slave at least as new as master).
As agreed with support, I have asked for an alert to be put into
the MySQL Network Monitoring and Advisory Service.
3) note that I'll re-enable rpl_insert_id as soon as 5.1-rpl gets
the changes from the main 5.1.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago  Bug#34043: Server loops excessively in _checkchunk() when safemalloc is enabled
Essentially, the problem is that safemalloc is excruciatingly
slow as it checks all allocated blocks for overrun at each
memory management primitive, yielding a almost exponential
slowdown for the memory management functions (malloc, realloc,
free). The overrun check basically consists of verifying some
bytes of a block for certain magic keys, which catches some
simple forms of overrun. Another minor problem is violation
of aliasing rules and that its own internal list of blocks
is prone to corruption.
Another issue with safemalloc is rather the maintenance cost
as the tool has a significant impact on the server code.
Given the magnitude of memory debuggers available nowadays,
especially those that are provided with the platform malloc
implementation, maintenance of a in-house and largely obsolete
memory debugger becomes a burden that is not worth the effort
due to its slowness and lack of support for detecting more
common forms of heap corruption.
Since there are third-party tools that can provide the same
functionality at a lower or comparable performance cost, the
solution is to simply remove safemalloc. Third-party tools
can provide the same functionality at a lower or comparable
performance cost.
The removal of safemalloc also allows a simplification of the
malloc wrappers, removing quite a bit of kludge: redefinition
of my_malloc, my_free and the removal of the unused second
argument of my_free. Since free() always check whether the
supplied pointer is null, redudant checks are also removed.
Also, this patch adds unit testing for my_malloc and moves
my_realloc implementation into the same file as the other
memory allocation primitives. 16 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago Manual merge from 5.0-rpl, of fixes for:
1)
BUG#25507 "multi-row insert delayed + auto increment causes
duplicate key entries on slave" (two concurrrent connections doing
multi-row INSERT DELAYED to insert into an auto_increment column,
caused replication slave to stop with "duplicate key error" (and
binlog was wrong), and BUG#26116 "If multi-row INSERT
DELAYED has errors, statement-based binlogging breaks" (the binlog
was not accounting for all rows inserted, or slave could stop).
The fix is that: in statement-based binlogging, a multi-row INSERT
DELAYED is silently converted to a non-delayed INSERT.
This is supposed to not affect many 5.1 users as in 5.1, the default
binlog format is "mixed", which does not have the bug (the bug is
only with binlog_format=STATEMENT).
We should document how the system delayed_insert thread decides of
its binlog format (which is not modified by this patch):
this decision is taken when the thread is created
and holds until it is terminated (is not affected by any later change
via SET GLOBAL BINLOG_FORMAT). It is also not affected by the binlog
format of the connection which issues INSERT DELAYED (this binlog
format does not affect how the row will be binlogged).
If one wants to change the binlog format of its server with SET
GLOBAL BINLOG_FORMAT, it should do FLUSH TABLES to be sure all
delayed_insert threads terminate and thus new threads are created,
taking into account the new format.
2)
BUG#24432
"INSERT... ON DUPLICATE KEY UPDATE skips auto_increment values".
When in an INSERT ON DUPLICATE KEY UPDATE, using
an autoincrement column, we inserted some autogenerated values and
also updated some rows, some autogenerated values were not used
(for example, even if 10 was the largest autoinc value in the table
at the start of the statement, 12 could be the first autogenerated
value inserted by the statement, instead of 11). One autogenerated
value was lost per updated row. Led to exhausting the range of the
autoincrement column faster.
Bug introduced by fix of BUG#20188; present since 5.0.24 and 5.1.12.
This bug breaks replication from a pre-5.0.24/pre-5.1.12 master.
But the present bugfix, as it makes INSERT ON DUP KEY UPDATE
behave like pre-5.0.24/pre-5.1.12, breaks replication from a
[5.0.24,5.0.34]/[5.1.12,5.1.15]
master to a fixed (5.0.36/5.1.16) slave! To warn users against this when
they upgrade their slave, as agreed with the support team, we add
code for a fixed slave to detect that it is connected to a buggy
master in a situation (INSERT ON DUP KEY UPDATE into autoinc column)
likely to break replication, in which case it cannot replicate so
stops and prints a message to the slave's error log and to SHOW SLAVE
STATUS.
For 5.0.36->[5.0.24,5.0.34] replication or 5.1.16->[5.1.12,5.1.15]
replication we cannot warn as master
does not know the slave's version (but we always recommended to users
to have slave at least as new as master).
As agreed with support, I have asked for an alert to be put into
the MySQL Network Monitoring and Advisory Service.
3) note that I'll re-enable rpl_insert_id as soon as 5.1-rpl gets
the changes from the main 5.1.
19 years ago Manual merge from 5.0-rpl, of fixes for:
1)
BUG#25507 "multi-row insert delayed + auto increment causes
duplicate key entries on slave" (two concurrrent connections doing
multi-row INSERT DELAYED to insert into an auto_increment column,
caused replication slave to stop with "duplicate key error" (and
binlog was wrong), and BUG#26116 "If multi-row INSERT
DELAYED has errors, statement-based binlogging breaks" (the binlog
was not accounting for all rows inserted, or slave could stop).
The fix is that: in statement-based binlogging, a multi-row INSERT
DELAYED is silently converted to a non-delayed INSERT.
This is supposed to not affect many 5.1 users as in 5.1, the default
binlog format is "mixed", which does not have the bug (the bug is
only with binlog_format=STATEMENT).
We should document how the system delayed_insert thread decides of
its binlog format (which is not modified by this patch):
this decision is taken when the thread is created
and holds until it is terminated (is not affected by any later change
via SET GLOBAL BINLOG_FORMAT). It is also not affected by the binlog
format of the connection which issues INSERT DELAYED (this binlog
format does not affect how the row will be binlogged).
If one wants to change the binlog format of its server with SET
GLOBAL BINLOG_FORMAT, it should do FLUSH TABLES to be sure all
delayed_insert threads terminate and thus new threads are created,
taking into account the new format.
2)
BUG#24432
"INSERT... ON DUPLICATE KEY UPDATE skips auto_increment values".
When in an INSERT ON DUPLICATE KEY UPDATE, using
an autoincrement column, we inserted some autogenerated values and
also updated some rows, some autogenerated values were not used
(for example, even if 10 was the largest autoinc value in the table
at the start of the statement, 12 could be the first autogenerated
value inserted by the statement, instead of 11). One autogenerated
value was lost per updated row. Led to exhausting the range of the
autoincrement column faster.
Bug introduced by fix of BUG#20188; present since 5.0.24 and 5.1.12.
This bug breaks replication from a pre-5.0.24/pre-5.1.12 master.
But the present bugfix, as it makes INSERT ON DUP KEY UPDATE
behave like pre-5.0.24/pre-5.1.12, breaks replication from a
[5.0.24,5.0.34]/[5.1.12,5.1.15]
master to a fixed (5.0.36/5.1.16) slave! To warn users against this when
they upgrade their slave, as agreed with the support team, we add
code for a fixed slave to detect that it is connected to a buggy
master in a situation (INSERT ON DUP KEY UPDATE into autoinc column)
likely to break replication, in which case it cannot replicate so
stops and prints a message to the slave's error log and to SHOW SLAVE
STATUS.
For 5.0.36->[5.0.24,5.0.34] replication or 5.1.16->[5.1.12,5.1.15]
replication we cannot warn as master
does not know the slave's version (but we always recommended to users
to have slave at least as new as master).
As agreed with support, I have asked for an alert to be put into
the MySQL Network Monitoring and Advisory Service.
3) note that I'll re-enable rpl_insert_id as soon as 5.1-rpl gets
the changes from the main 5.1.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago Manual merge from 5.0-rpl, of fixes for:
1)
BUG#25507 "multi-row insert delayed + auto increment causes
duplicate key entries on slave" (two concurrrent connections doing
multi-row INSERT DELAYED to insert into an auto_increment column,
caused replication slave to stop with "duplicate key error" (and
binlog was wrong), and BUG#26116 "If multi-row INSERT
DELAYED has errors, statement-based binlogging breaks" (the binlog
was not accounting for all rows inserted, or slave could stop).
The fix is that: in statement-based binlogging, a multi-row INSERT
DELAYED is silently converted to a non-delayed INSERT.
This is supposed to not affect many 5.1 users as in 5.1, the default
binlog format is "mixed", which does not have the bug (the bug is
only with binlog_format=STATEMENT).
We should document how the system delayed_insert thread decides of
its binlog format (which is not modified by this patch):
this decision is taken when the thread is created
and holds until it is terminated (is not affected by any later change
via SET GLOBAL BINLOG_FORMAT). It is also not affected by the binlog
format of the connection which issues INSERT DELAYED (this binlog
format does not affect how the row will be binlogged).
If one wants to change the binlog format of its server with SET
GLOBAL BINLOG_FORMAT, it should do FLUSH TABLES to be sure all
delayed_insert threads terminate and thus new threads are created,
taking into account the new format.
2)
BUG#24432
"INSERT... ON DUPLICATE KEY UPDATE skips auto_increment values".
When in an INSERT ON DUPLICATE KEY UPDATE, using
an autoincrement column, we inserted some autogenerated values and
also updated some rows, some autogenerated values were not used
(for example, even if 10 was the largest autoinc value in the table
at the start of the statement, 12 could be the first autogenerated
value inserted by the statement, instead of 11). One autogenerated
value was lost per updated row. Led to exhausting the range of the
autoincrement column faster.
Bug introduced by fix of BUG#20188; present since 5.0.24 and 5.1.12.
This bug breaks replication from a pre-5.0.24/pre-5.1.12 master.
But the present bugfix, as it makes INSERT ON DUP KEY UPDATE
behave like pre-5.0.24/pre-5.1.12, breaks replication from a
[5.0.24,5.0.34]/[5.1.12,5.1.15]
master to a fixed (5.0.36/5.1.16) slave! To warn users against this when
they upgrade their slave, as agreed with the support team, we add
code for a fixed slave to detect that it is connected to a buggy
master in a situation (INSERT ON DUP KEY UPDATE into autoinc column)
likely to break replication, in which case it cannot replicate so
stops and prints a message to the slave's error log and to SHOW SLAVE
STATUS.
For 5.0.36->[5.0.24,5.0.34] replication or 5.1.16->[5.1.12,5.1.15]
replication we cannot warn as master
does not know the slave's version (but we always recommended to users
to have slave at least as new as master).
As agreed with support, I have asked for an alert to be put into
the MySQL Network Monitoring and Advisory Service.
3) note that I'll re-enable rpl_insert_id as soon as 5.1-rpl gets
the changes from the main 5.1.
19 years ago  BUG#37426: RBR breaks for CHAR() UTF-8 fields > 85 chars
In order to handle CHAR() fields, 8 bits were reserved for
the size of the CHAR field. However, instead of denoting the
number of characters in the field, field_length was used which
denotes the number of bytes in the field.
Since UTF-8 fields can have three bytes per character (and
has been extended to have four bytes per character in 6.0),
an extra two bits have been encoded in the field metadata
work for fields of type Field_string (i.e., CHAR fields).
Since the metadata word is filled, the extra bits have been
encoded in the upper 4 bits of the real type (the most
significant byte of the metadata word) by computing the
bitwise xor of the extra two bits. Since the upper 4 bits
of the real type always is 1111 for Field_string, this
means that for fields of length <256, the encoding is
identical to the encoding used in pre-5.1.26 servers, but
for lengths of 256 or more, an unrecognized type is formed,
causing an old slave (that does not handle lengths of 256
or more) to stop.
18 years ago Manual merge from 5.0-rpl, of fixes for:
1)
BUG#25507 "multi-row insert delayed + auto increment causes
duplicate key entries on slave" (two concurrrent connections doing
multi-row INSERT DELAYED to insert into an auto_increment column,
caused replication slave to stop with "duplicate key error" (and
binlog was wrong), and BUG#26116 "If multi-row INSERT
DELAYED has errors, statement-based binlogging breaks" (the binlog
was not accounting for all rows inserted, or slave could stop).
The fix is that: in statement-based binlogging, a multi-row INSERT
DELAYED is silently converted to a non-delayed INSERT.
This is supposed to not affect many 5.1 users as in 5.1, the default
binlog format is "mixed", which does not have the bug (the bug is
only with binlog_format=STATEMENT).
We should document how the system delayed_insert thread decides of
its binlog format (which is not modified by this patch):
this decision is taken when the thread is created
and holds until it is terminated (is not affected by any later change
via SET GLOBAL BINLOG_FORMAT). It is also not affected by the binlog
format of the connection which issues INSERT DELAYED (this binlog
format does not affect how the row will be binlogged).
If one wants to change the binlog format of its server with SET
GLOBAL BINLOG_FORMAT, it should do FLUSH TABLES to be sure all
delayed_insert threads terminate and thus new threads are created,
taking into account the new format.
2)
BUG#24432
"INSERT... ON DUPLICATE KEY UPDATE skips auto_increment values".
When in an INSERT ON DUPLICATE KEY UPDATE, using
an autoincrement column, we inserted some autogenerated values and
also updated some rows, some autogenerated values were not used
(for example, even if 10 was the largest autoinc value in the table
at the start of the statement, 12 could be the first autogenerated
value inserted by the statement, instead of 11). One autogenerated
value was lost per updated row. Led to exhausting the range of the
autoincrement column faster.
Bug introduced by fix of BUG#20188; present since 5.0.24 and 5.1.12.
This bug breaks replication from a pre-5.0.24/pre-5.1.12 master.
But the present bugfix, as it makes INSERT ON DUP KEY UPDATE
behave like pre-5.0.24/pre-5.1.12, breaks replication from a
[5.0.24,5.0.34]/[5.1.12,5.1.15]
master to a fixed (5.0.36/5.1.16) slave! To warn users against this when
they upgrade their slave, as agreed with the support team, we add
code for a fixed slave to detect that it is connected to a buggy
master in a situation (INSERT ON DUP KEY UPDATE into autoinc column)
likely to break replication, in which case it cannot replicate so
stops and prints a message to the slave's error log and to SHOW SLAVE
STATUS.
For 5.0.36->[5.0.24,5.0.34] replication or 5.1.16->[5.1.12,5.1.15]
replication we cannot warn as master
does not know the slave's version (but we always recommended to users
to have slave at least as new as master).
As agreed with support, I have asked for an alert to be put into
the MySQL Network Monitoring and Advisory Service.
3) note that I'll re-enable rpl_insert_id as soon as 5.1-rpl gets
the changes from the main 5.1.
19 years ago WL#3146 "less locking in auto_increment":
this is a cleanup patch for our current auto_increment handling:
new names for auto_increment variables in THD, new methods to manipulate them
(see sql_class.h), some move into handler::, causing less backup/restore
work when executing substatements.
This makes the logic hopefully clearer, less work is is needed in
mysql_insert().
By cleaning up, using different variables for different purposes (instead
of one for 3 things...), we fix those bugs, which someone may want to fix
in 5.0 too:
BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate
statement-based"
BUG#20341 "stored function inserting into one auto_increment puts bad
data in slave"
BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE"
(now if a row is updated, LAST_INSERT_ID() will return its id)
and re-fixes:
BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT"
(already fixed differently by Ramil in 4.1)
Test of documented behaviour of mysql_insert_id() (there was no test).
The behaviour changes introduced are:
- LAST_INSERT_ID() now returns "the first autogenerated auto_increment value
successfully inserted", instead of "the first autogenerated auto_increment
value if any row was successfully inserted", see auto_increment.test.
Same for mysql_insert_id(), see mysql_client_test.c.
- LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY
UPDATE, see auto_increment.test. Same for mysql_insert_id(), see
mysql_client_test.c.
- LAST_INSERT_ID() does not change if no autogenerated value was successfully
inserted (it used to then be 0), see auto_increment.test.
- if in INSERT SELECT no autogenerated value was successfully inserted,
mysql_insert_id() now returns the id of the last inserted row (it already
did this for INSERT VALUES), see mysql_client_test.c.
- if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X
(it already did this for INSERT VALUES), see mysql_client_test.c.
- NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE,
the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID
influences not only the first row now.
Additionally, when unlocking a table we check that the thread is not keeping
a next_insert_id (as the table is unlocked that id is potentially out-of-date);
forgetting about this next_insert_id is done in a new
handler::ha_release_auto_increment().
Finally we prepare for engines capable of reserving finite-length intervals
of auto_increment values: we store such intervals in THD. The next step
(to be done by the replication team in 5.1) is to read those intervals from
THD and actually store them in the statement-based binary log. NDB
will be a good engine to test that.
20 years ago WL#3146 "less locking in auto_increment":
this is a cleanup patch for our current auto_increment handling:
new names for auto_increment variables in THD, new methods to manipulate them
(see sql_class.h), some move into handler::, causing less backup/restore
work when executing substatements.
This makes the logic hopefully clearer, less work is is needed in
mysql_insert().
By cleaning up, using different variables for different purposes (instead
of one for 3 things...), we fix those bugs, which someone may want to fix
in 5.0 too:
BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate
statement-based"
BUG#20341 "stored function inserting into one auto_increment puts bad
data in slave"
BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE"
(now if a row is updated, LAST_INSERT_ID() will return its id)
and re-fixes:
BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT"
(already fixed differently by Ramil in 4.1)
Test of documented behaviour of mysql_insert_id() (there was no test).
The behaviour changes introduced are:
- LAST_INSERT_ID() now returns "the first autogenerated auto_increment value
successfully inserted", instead of "the first autogenerated auto_increment
value if any row was successfully inserted", see auto_increment.test.
Same for mysql_insert_id(), see mysql_client_test.c.
- LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY
UPDATE, see auto_increment.test. Same for mysql_insert_id(), see
mysql_client_test.c.
- LAST_INSERT_ID() does not change if no autogenerated value was successfully
inserted (it used to then be 0), see auto_increment.test.
- if in INSERT SELECT no autogenerated value was successfully inserted,
mysql_insert_id() now returns the id of the last inserted row (it already
did this for INSERT VALUES), see mysql_client_test.c.
- if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X
(it already did this for INSERT VALUES), see mysql_client_test.c.
- NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE,
the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID
influences not only the first row now.
Additionally, when unlocking a table we check that the thread is not keeping
a next_insert_id (as the table is unlocked that id is potentially out-of-date);
forgetting about this next_insert_id is done in a new
handler::ha_release_auto_increment().
Finally we prepare for engines capable of reserving finite-length intervals
of auto_increment values: we store such intervals in THD. The next step
(to be done by the replication team in 5.1) is to read those intervals from
THD and actually store them in the statement-based binary log. NDB
will be a good engine to test that.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago WL#3146 "less locking in auto_increment":
this is a cleanup patch for our current auto_increment handling:
new names for auto_increment variables in THD, new methods to manipulate them
(see sql_class.h), some move into handler::, causing less backup/restore
work when executing substatements.
This makes the logic hopefully clearer, less work is is needed in
mysql_insert().
By cleaning up, using different variables for different purposes (instead
of one for 3 things...), we fix those bugs, which someone may want to fix
in 5.0 too:
BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate
statement-based"
BUG#20341 "stored function inserting into one auto_increment puts bad
data in slave"
BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE"
(now if a row is updated, LAST_INSERT_ID() will return its id)
and re-fixes:
BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT"
(already fixed differently by Ramil in 4.1)
Test of documented behaviour of mysql_insert_id() (there was no test).
The behaviour changes introduced are:
- LAST_INSERT_ID() now returns "the first autogenerated auto_increment value
successfully inserted", instead of "the first autogenerated auto_increment
value if any row was successfully inserted", see auto_increment.test.
Same for mysql_insert_id(), see mysql_client_test.c.
- LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY
UPDATE, see auto_increment.test. Same for mysql_insert_id(), see
mysql_client_test.c.
- LAST_INSERT_ID() does not change if no autogenerated value was successfully
inserted (it used to then be 0), see auto_increment.test.
- if in INSERT SELECT no autogenerated value was successfully inserted,
mysql_insert_id() now returns the id of the last inserted row (it already
did this for INSERT VALUES), see mysql_client_test.c.
- if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X
(it already did this for INSERT VALUES), see mysql_client_test.c.
- NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE,
the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID
influences not only the first row now.
Additionally, when unlocking a table we check that the thread is not keeping
a next_insert_id (as the table is unlocked that id is potentially out-of-date);
forgetting about this next_insert_id is done in a new
handler::ha_release_auto_increment().
Finally we prepare for engines capable of reserving finite-length intervals
of auto_increment values: we store such intervals in THD. The next step
(to be done by the replication team in 5.1) is to read those intervals from
THD and actually store them in the statement-based binary log. NDB
will be a good engine to test that.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago WL#3146 "less locking in auto_increment":
this is a cleanup patch for our current auto_increment handling:
new names for auto_increment variables in THD, new methods to manipulate them
(see sql_class.h), some move into handler::, causing less backup/restore
work when executing substatements.
This makes the logic hopefully clearer, less work is is needed in
mysql_insert().
By cleaning up, using different variables for different purposes (instead
of one for 3 things...), we fix those bugs, which someone may want to fix
in 5.0 too:
BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate
statement-based"
BUG#20341 "stored function inserting into one auto_increment puts bad
data in slave"
BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE"
(now if a row is updated, LAST_INSERT_ID() will return its id)
and re-fixes:
BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT"
(already fixed differently by Ramil in 4.1)
Test of documented behaviour of mysql_insert_id() (there was no test).
The behaviour changes introduced are:
- LAST_INSERT_ID() now returns "the first autogenerated auto_increment value
successfully inserted", instead of "the first autogenerated auto_increment
value if any row was successfully inserted", see auto_increment.test.
Same for mysql_insert_id(), see mysql_client_test.c.
- LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY
UPDATE, see auto_increment.test. Same for mysql_insert_id(), see
mysql_client_test.c.
- LAST_INSERT_ID() does not change if no autogenerated value was successfully
inserted (it used to then be 0), see auto_increment.test.
- if in INSERT SELECT no autogenerated value was successfully inserted,
mysql_insert_id() now returns the id of the last inserted row (it already
did this for INSERT VALUES), see mysql_client_test.c.
- if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X
(it already did this for INSERT VALUES), see mysql_client_test.c.
- NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE,
the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID
influences not only the first row now.
Additionally, when unlocking a table we check that the thread is not keeping
a next_insert_id (as the table is unlocked that id is potentially out-of-date);
forgetting about this next_insert_id is done in a new
handler::ha_release_auto_increment().
Finally we prepare for engines capable of reserving finite-length intervals
of auto_increment values: we store such intervals in THD. The next step
(to be done by the replication team in 5.1) is to read those intervals from
THD and actually store them in the statement-based binary log. NDB
will be a good engine to test that.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago WL#3146 "less locking in auto_increment":
this is a cleanup patch for our current auto_increment handling:
new names for auto_increment variables in THD, new methods to manipulate them
(see sql_class.h), some move into handler::, causing less backup/restore
work when executing substatements.
This makes the logic hopefully clearer, less work is is needed in
mysql_insert().
By cleaning up, using different variables for different purposes (instead
of one for 3 things...), we fix those bugs, which someone may want to fix
in 5.0 too:
BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate
statement-based"
BUG#20341 "stored function inserting into one auto_increment puts bad
data in slave"
BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE"
(now if a row is updated, LAST_INSERT_ID() will return its id)
and re-fixes:
BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT"
(already fixed differently by Ramil in 4.1)
Test of documented behaviour of mysql_insert_id() (there was no test).
The behaviour changes introduced are:
- LAST_INSERT_ID() now returns "the first autogenerated auto_increment value
successfully inserted", instead of "the first autogenerated auto_increment
value if any row was successfully inserted", see auto_increment.test.
Same for mysql_insert_id(), see mysql_client_test.c.
- LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY
UPDATE, see auto_increment.test. Same for mysql_insert_id(), see
mysql_client_test.c.
- LAST_INSERT_ID() does not change if no autogenerated value was successfully
inserted (it used to then be 0), see auto_increment.test.
- if in INSERT SELECT no autogenerated value was successfully inserted,
mysql_insert_id() now returns the id of the last inserted row (it already
did this for INSERT VALUES), see mysql_client_test.c.
- if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X
(it already did this for INSERT VALUES), see mysql_client_test.c.
- NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE,
the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID
influences not only the first row now.
Additionally, when unlocking a table we check that the thread is not keeping
a next_insert_id (as the table is unlocked that id is potentially out-of-date);
forgetting about this next_insert_id is done in a new
handler::ha_release_auto_increment().
Finally we prepare for engines capable of reserving finite-length intervals
of auto_increment values: we store such intervals in THD. The next step
(to be done by the replication team in 5.1) is to read those intervals from
THD and actually store them in the statement-based binary log. NDB
will be a good engine to test that.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
19 years ago Manual merge from 5.0-rpl, of fixes for:
1)
BUG#25507 "multi-row insert delayed + auto increment causes
duplicate key entries on slave" (two concurrrent connections doing
multi-row INSERT DELAYED to insert into an auto_increment column,
caused replication slave to stop with "duplicate key error" (and
binlog was wrong), and BUG#26116 "If multi-row INSERT
DELAYED has errors, statement-based binlogging breaks" (the binlog
was not accounting for all rows inserted, or slave could stop).
The fix is that: in statement-based binlogging, a multi-row INSERT
DELAYED is silently converted to a non-delayed INSERT.
This is supposed to not affect many 5.1 users as in 5.1, the default
binlog format is "mixed", which does not have the bug (the bug is
only with binlog_format=STATEMENT).
We should document how the system delayed_insert thread decides of
its binlog format (which is not modified by this patch):
this decision is taken when the thread is created
and holds until it is terminated (is not affected by any later change
via SET GLOBAL BINLOG_FORMAT). It is also not affected by the binlog
format of the connection which issues INSERT DELAYED (this binlog
format does not affect how the row will be binlogged).
If one wants to change the binlog format of its server with SET
GLOBAL BINLOG_FORMAT, it should do FLUSH TABLES to be sure all
delayed_insert threads terminate and thus new threads are created,
taking into account the new format.
2)
BUG#24432
"INSERT... ON DUPLICATE KEY UPDATE skips auto_increment values".
When in an INSERT ON DUPLICATE KEY UPDATE, using
an autoincrement column, we inserted some autogenerated values and
also updated some rows, some autogenerated values were not used
(for example, even if 10 was the largest autoinc value in the table
at the start of the statement, 12 could be the first autogenerated
value inserted by the statement, instead of 11). One autogenerated
value was lost per updated row. Led to exhausting the range of the
autoincrement column faster.
Bug introduced by fix of BUG#20188; present since 5.0.24 and 5.1.12.
This bug breaks replication from a pre-5.0.24/pre-5.1.12 master.
But the present bugfix, as it makes INSERT ON DUP KEY UPDATE
behave like pre-5.0.24/pre-5.1.12, breaks replication from a
[5.0.24,5.0.34]/[5.1.12,5.1.15]
master to a fixed (5.0.36/5.1.16) slave! To warn users against this when
they upgrade their slave, as agreed with the support team, we add
code for a fixed slave to detect that it is connected to a buggy
master in a situation (INSERT ON DUP KEY UPDATE into autoinc column)
likely to break replication, in which case it cannot replicate so
stops and prints a message to the slave's error log and to SHOW SLAVE
STATUS.
For 5.0.36->[5.0.24,5.0.34] replication or 5.1.16->[5.1.12,5.1.15]
replication we cannot warn as master
does not know the slave's version (but we always recommended to users
to have slave at least as new as master).
As agreed with support, I have asked for an alert to be put into
the MySQL Network Monitoring and Advisory Service.
3) note that I'll re-enable rpl_insert_id as soon as 5.1-rpl gets
the changes from the main 5.1.
19 years ago WL#3146 "less locking in auto_increment":
this is a cleanup patch for our current auto_increment handling:
new names for auto_increment variables in THD, new methods to manipulate them
(see sql_class.h), some move into handler::, causing less backup/restore
work when executing substatements.
This makes the logic hopefully clearer, less work is is needed in
mysql_insert().
By cleaning up, using different variables for different purposes (instead
of one for 3 things...), we fix those bugs, which someone may want to fix
in 5.0 too:
BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate
statement-based"
BUG#20341 "stored function inserting into one auto_increment puts bad
data in slave"
BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE"
(now if a row is updated, LAST_INSERT_ID() will return its id)
and re-fixes:
BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT"
(already fixed differently by Ramil in 4.1)
Test of documented behaviour of mysql_insert_id() (there was no test).
The behaviour changes introduced are:
- LAST_INSERT_ID() now returns "the first autogenerated auto_increment value
successfully inserted", instead of "the first autogenerated auto_increment
value if any row was successfully inserted", see auto_increment.test.
Same for mysql_insert_id(), see mysql_client_test.c.
- LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY
UPDATE, see auto_increment.test. Same for mysql_insert_id(), see
mysql_client_test.c.
- LAST_INSERT_ID() does not change if no autogenerated value was successfully
inserted (it used to then be 0), see auto_increment.test.
- if in INSERT SELECT no autogenerated value was successfully inserted,
mysql_insert_id() now returns the id of the last inserted row (it already
did this for INSERT VALUES), see mysql_client_test.c.
- if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X
(it already did this for INSERT VALUES), see mysql_client_test.c.
- NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE,
the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID
influences not only the first row now.
Additionally, when unlocking a table we check that the thread is not keeping
a next_insert_id (as the table is unlocked that id is potentially out-of-date);
forgetting about this next_insert_id is done in a new
handler::ha_release_auto_increment().
Finally we prepare for engines capable of reserving finite-length intervals
of auto_increment values: we store such intervals in THD. The next step
(to be done by the replication team in 5.1) is to read those intervals from
THD and actually store them in the statement-based binary log. NDB
will be a good engine to test that.
20 years ago WL#3146 "less locking in auto_increment":
this is a cleanup patch for our current auto_increment handling:
new names for auto_increment variables in THD, new methods to manipulate them
(see sql_class.h), some move into handler::, causing less backup/restore
work when executing substatements.
This makes the logic hopefully clearer, less work is is needed in
mysql_insert().
By cleaning up, using different variables for different purposes (instead
of one for 3 things...), we fix those bugs, which someone may want to fix
in 5.0 too:
BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate
statement-based"
BUG#20341 "stored function inserting into one auto_increment puts bad
data in slave"
BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE"
(now if a row is updated, LAST_INSERT_ID() will return its id)
and re-fixes:
BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT"
(already fixed differently by Ramil in 4.1)
Test of documented behaviour of mysql_insert_id() (there was no test).
The behaviour changes introduced are:
- LAST_INSERT_ID() now returns "the first autogenerated auto_increment value
successfully inserted", instead of "the first autogenerated auto_increment
value if any row was successfully inserted", see auto_increment.test.
Same for mysql_insert_id(), see mysql_client_test.c.
- LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY
UPDATE, see auto_increment.test. Same for mysql_insert_id(), see
mysql_client_test.c.
- LAST_INSERT_ID() does not change if no autogenerated value was successfully
inserted (it used to then be 0), see auto_increment.test.
- if in INSERT SELECT no autogenerated value was successfully inserted,
mysql_insert_id() now returns the id of the last inserted row (it already
did this for INSERT VALUES), see mysql_client_test.c.
- if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X
(it already did this for INSERT VALUES), see mysql_client_test.c.
- NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE,
the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID
influences not only the first row now.
Additionally, when unlocking a table we check that the thread is not keeping
a next_insert_id (as the table is unlocked that id is potentially out-of-date);
forgetting about this next_insert_id is done in a new
handler::ha_release_auto_increment().
Finally we prepare for engines capable of reserving finite-length intervals
of auto_increment values: we store such intervals in THD. The next step
(to be done by the replication team in 5.1) is to read those intervals from
THD and actually store them in the statement-based binary log. NDB
will be a good engine to test that.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago WL#3146 "less locking in auto_increment":
this is a cleanup patch for our current auto_increment handling:
new names for auto_increment variables in THD, new methods to manipulate them
(see sql_class.h), some move into handler::, causing less backup/restore
work when executing substatements.
This makes the logic hopefully clearer, less work is is needed in
mysql_insert().
By cleaning up, using different variables for different purposes (instead
of one for 3 things...), we fix those bugs, which someone may want to fix
in 5.0 too:
BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate
statement-based"
BUG#20341 "stored function inserting into one auto_increment puts bad
data in slave"
BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE"
(now if a row is updated, LAST_INSERT_ID() will return its id)
and re-fixes:
BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT"
(already fixed differently by Ramil in 4.1)
Test of documented behaviour of mysql_insert_id() (there was no test).
The behaviour changes introduced are:
- LAST_INSERT_ID() now returns "the first autogenerated auto_increment value
successfully inserted", instead of "the first autogenerated auto_increment
value if any row was successfully inserted", see auto_increment.test.
Same for mysql_insert_id(), see mysql_client_test.c.
- LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY
UPDATE, see auto_increment.test. Same for mysql_insert_id(), see
mysql_client_test.c.
- LAST_INSERT_ID() does not change if no autogenerated value was successfully
inserted (it used to then be 0), see auto_increment.test.
- if in INSERT SELECT no autogenerated value was successfully inserted,
mysql_insert_id() now returns the id of the last inserted row (it already
did this for INSERT VALUES), see mysql_client_test.c.
- if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X
(it already did this for INSERT VALUES), see mysql_client_test.c.
- NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE,
the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID
influences not only the first row now.
Additionally, when unlocking a table we check that the thread is not keeping
a next_insert_id (as the table is unlocked that id is potentially out-of-date);
forgetting about this next_insert_id is done in a new
handler::ha_release_auto_increment().
Finally we prepare for engines capable of reserving finite-length intervals
of auto_increment values: we store such intervals in THD. The next step
(to be done by the replication team in 5.1) is to read those intervals from
THD and actually store them in the statement-based binary log. NDB
will be a good engine to test that.
20 years ago WL#3146 "less locking in auto_increment":
this is a cleanup patch for our current auto_increment handling:
new names for auto_increment variables in THD, new methods to manipulate them
(see sql_class.h), some move into handler::, causing less backup/restore
work when executing substatements.
This makes the logic hopefully clearer, less work is is needed in
mysql_insert().
By cleaning up, using different variables for different purposes (instead
of one for 3 things...), we fix those bugs, which someone may want to fix
in 5.0 too:
BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate
statement-based"
BUG#20341 "stored function inserting into one auto_increment puts bad
data in slave"
BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE"
(now if a row is updated, LAST_INSERT_ID() will return its id)
and re-fixes:
BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT"
(already fixed differently by Ramil in 4.1)
Test of documented behaviour of mysql_insert_id() (there was no test).
The behaviour changes introduced are:
- LAST_INSERT_ID() now returns "the first autogenerated auto_increment value
successfully inserted", instead of "the first autogenerated auto_increment
value if any row was successfully inserted", see auto_increment.test.
Same for mysql_insert_id(), see mysql_client_test.c.
- LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY
UPDATE, see auto_increment.test. Same for mysql_insert_id(), see
mysql_client_test.c.
- LAST_INSERT_ID() does not change if no autogenerated value was successfully
inserted (it used to then be 0), see auto_increment.test.
- if in INSERT SELECT no autogenerated value was successfully inserted,
mysql_insert_id() now returns the id of the last inserted row (it already
did this for INSERT VALUES), see mysql_client_test.c.
- if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X
(it already did this for INSERT VALUES), see mysql_client_test.c.
- NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE,
the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID
influences not only the first row now.
Additionally, when unlocking a table we check that the thread is not keeping
a next_insert_id (as the table is unlocked that id is potentially out-of-date);
forgetting about this next_insert_id is done in a new
handler::ha_release_auto_increment().
Finally we prepare for engines capable of reserving finite-length intervals
of auto_increment values: we store such intervals in THD. The next step
(to be done by the replication team in 5.1) is to read those intervals from
THD and actually store them in the statement-based binary log. NDB
will be a good engine to test that.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago WL#3146 "less locking in auto_increment":
this is a cleanup patch for our current auto_increment handling:
new names for auto_increment variables in THD, new methods to manipulate them
(see sql_class.h), some move into handler::, causing less backup/restore
work when executing substatements.
This makes the logic hopefully clearer, less work is is needed in
mysql_insert().
By cleaning up, using different variables for different purposes (instead
of one for 3 things...), we fix those bugs, which someone may want to fix
in 5.0 too:
BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate
statement-based"
BUG#20341 "stored function inserting into one auto_increment puts bad
data in slave"
BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE"
(now if a row is updated, LAST_INSERT_ID() will return its id)
and re-fixes:
BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT"
(already fixed differently by Ramil in 4.1)
Test of documented behaviour of mysql_insert_id() (there was no test).
The behaviour changes introduced are:
- LAST_INSERT_ID() now returns "the first autogenerated auto_increment value
successfully inserted", instead of "the first autogenerated auto_increment
value if any row was successfully inserted", see auto_increment.test.
Same for mysql_insert_id(), see mysql_client_test.c.
- LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY
UPDATE, see auto_increment.test. Same for mysql_insert_id(), see
mysql_client_test.c.
- LAST_INSERT_ID() does not change if no autogenerated value was successfully
inserted (it used to then be 0), see auto_increment.test.
- if in INSERT SELECT no autogenerated value was successfully inserted,
mysql_insert_id() now returns the id of the last inserted row (it already
did this for INSERT VALUES), see mysql_client_test.c.
- if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X
(it already did this for INSERT VALUES), see mysql_client_test.c.
- NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE,
the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID
influences not only the first row now.
Additionally, when unlocking a table we check that the thread is not keeping
a next_insert_id (as the table is unlocked that id is potentially out-of-date);
forgetting about this next_insert_id is done in a new
handler::ha_release_auto_increment().
Finally we prepare for engines capable of reserving finite-length intervals
of auto_increment values: we store such intervals in THD. The next step
(to be done by the replication team in 5.1) is to read those intervals from
THD and actually store them in the statement-based binary log. NDB
will be a good engine to test that.
20 years ago Bug#34043: Server loops excessively in _checkchunk() when safemalloc is enabled
Essentially, the problem is that safemalloc is excruciatingly
slow as it checks all allocated blocks for overrun at each
memory management primitive, yielding a almost exponential
slowdown for the memory management functions (malloc, realloc,
free). The overrun check basically consists of verifying some
bytes of a block for certain magic keys, which catches some
simple forms of overrun. Another minor problem is violation
of aliasing rules and that its own internal list of blocks
is prone to corruption.
Another issue with safemalloc is rather the maintenance cost
as the tool has a significant impact on the server code.
Given the magnitude of memory debuggers available nowadays,
especially those that are provided with the platform malloc
implementation, maintenance of a in-house and largely obsolete
memory debugger becomes a burden that is not worth the effort
due to its slowness and lack of support for detecting more
common forms of heap corruption.
Since there are third-party tools that can provide the same
functionality at a lower or comparable performance cost, the
solution is to simply remove safemalloc. Third-party tools
can provide the same functionality at a lower or comparable
performance cost.
The removal of safemalloc also allows a simplification of the
malloc wrappers, removing quite a bit of kludge: redefinition
of my_malloc, my_free and the removal of the unused second
argument of my_free. Since free() always check whether the
supplied pointer is null, redudant checks are also removed.
Also, this patch adds unit testing for my_malloc and moves
my_realloc implementation into the same file as the other
memory allocation primitives. 16 years ago  BUG#39934: Slave stops for engine that only support row-based logging
General overview:
The logic for switching to row format when binlog_format=MIXED had
numerous flaws. The underlying problem was the lack of a consistent
architecture.
General purpose of this changeset:
This changeset introduces an architecture for switching to row format
when binlog_format=MIXED. It enforces the architecture where it has
to. It leaves some bugs to be fixed later. It adds extensive tests to
verify that unsafe statements work as expected and that appropriate
errors are produced by problems with the selection of binlog format.
It was not practical to split this into smaller pieces of work.
Problem 1:
To determine the logging mode, the code has to take several parameters
into account (namely: (1) the value of binlog_format; (2) the
capabilities of the engines; (3) the type of the current statement:
normal, unsafe, or row injection). These parameters may conflict in
several ways, namely:
- binlog_format=STATEMENT for a row injection
- binlog_format=STATEMENT for an unsafe statement
- binlog_format=STATEMENT for an engine only supporting row logging
- binlog_format=ROW for an engine only supporting statement logging
- statement is unsafe and engine does not support row logging
- row injection in a table that does not support statement logging
- statement modifies one table that does not support row logging and
one that does not support statement logging
Several of these conflicts were not detected, or were detected with
an inappropriate error message. The problem of BUG#39934 was that no
appropriate error message was written for the case when an engine
only supporting row logging executed a row injection with
binlog_format=ROW. However, all above cases must be handled.
Fix 1:
Introduce new error codes (sql/share/errmsg.txt). Ensure that all
conditions are detected and handled in decide_logging_format()
Problem 2:
The binlog format shall be determined once per statement, in
decide_logging_format(). It shall not be changed before or after that.
Before decide_logging_format() is called, all information necessary to
determine the logging format must be available. This principle ensures
that all unsafe statements are handled in a consistent way.
However, this principle is not followed:
thd->set_current_stmt_binlog_row_based_if_mixed() is called in several
places, including from code executing UPDATE..LIMIT,
INSERT..SELECT..LIMIT, DELETE..LIMIT, INSERT DELAYED, and
SET @@binlog_format. After Problem 1 was fixed, that caused
inconsistencies where these unsafe statements would not print the
appropriate warnings or errors for some of the conflicts.
Fix 2:
Remove calls to THD::set_current_stmt_binlog_row_based_if_mixed() from
code executed after decide_logging_format(). Compensate by calling the
set_current_stmt_unsafe() at parse time. This way, all unsafe statements
are detected by decide_logging_format().
Problem 3:
INSERT DELAYED is not unsafe: it is logged in statement format even if
binlog_format=MIXED, and no warning is printed even if
binlog_format=STATEMENT. This is BUG#45825.
Fix 3:
Made INSERT DELAYED set itself to unsafe at parse time. This allows
decide_logging_format() to detect that a warning should be printed or
the binlog_format changed.
Problem 4:
LIMIT clause were not marked as unsafe when executed inside stored
functions/triggers/views/prepared statements. This is
BUG#45785.
Fix 4:
Make statements containing the LIMIT clause marked as unsafe at
parse time, instead of at execution time. This allows propagating
unsafe-ness to the view.
17 years ago BUG#39934: Slave stops for engine that only support row-based logging
General overview:
The logic for switching to row format when binlog_format=MIXED had
numerous flaws. The underlying problem was the lack of a consistent
architecture.
General purpose of this changeset:
This changeset introduces an architecture for switching to row format
when binlog_format=MIXED. It enforces the architecture where it has
to. It leaves some bugs to be fixed later. It adds extensive tests to
verify that unsafe statements work as expected and that appropriate
errors are produced by problems with the selection of binlog format.
It was not practical to split this into smaller pieces of work.
Problem 1:
To determine the logging mode, the code has to take several parameters
into account (namely: (1) the value of binlog_format; (2) the
capabilities of the engines; (3) the type of the current statement:
normal, unsafe, or row injection). These parameters may conflict in
several ways, namely:
- binlog_format=STATEMENT for a row injection
- binlog_format=STATEMENT for an unsafe statement
- binlog_format=STATEMENT for an engine only supporting row logging
- binlog_format=ROW for an engine only supporting statement logging
- statement is unsafe and engine does not support row logging
- row injection in a table that does not support statement logging
- statement modifies one table that does not support row logging and
one that does not support statement logging
Several of these conflicts were not detected, or were detected with
an inappropriate error message. The problem of BUG#39934 was that no
appropriate error message was written for the case when an engine
only supporting row logging executed a row injection with
binlog_format=ROW. However, all above cases must be handled.
Fix 1:
Introduce new error codes (sql/share/errmsg.txt). Ensure that all
conditions are detected and handled in decide_logging_format()
Problem 2:
The binlog format shall be determined once per statement, in
decide_logging_format(). It shall not be changed before or after that.
Before decide_logging_format() is called, all information necessary to
determine the logging format must be available. This principle ensures
that all unsafe statements are handled in a consistent way.
However, this principle is not followed:
thd->set_current_stmt_binlog_row_based_if_mixed() is called in several
places, including from code executing UPDATE..LIMIT,
INSERT..SELECT..LIMIT, DELETE..LIMIT, INSERT DELAYED, and
SET @@binlog_format. After Problem 1 was fixed, that caused
inconsistencies where these unsafe statements would not print the
appropriate warnings or errors for some of the conflicts.
Fix 2:
Remove calls to THD::set_current_stmt_binlog_row_based_if_mixed() from
code executed after decide_logging_format(). Compensate by calling the
set_current_stmt_unsafe() at parse time. This way, all unsafe statements
are detected by decide_logging_format().
Problem 3:
INSERT DELAYED is not unsafe: it is logged in statement format even if
binlog_format=MIXED, and no warning is printed even if
binlog_format=STATEMENT. This is BUG#45825.
Fix 3:
Made INSERT DELAYED set itself to unsafe at parse time. This allows
decide_logging_format() to detect that a warning should be printed or
the binlog_format changed.
Problem 4:
LIMIT clause were not marked as unsafe when executed inside stored
functions/triggers/views/prepared statements. This is
BUG#45785.
Fix 4:
Make statements containing the LIMIT clause marked as unsafe at
parse time, instead of at execution time. This allows propagating
unsafe-ness to the view.
17 years ago BUG#39934: Slave stops for engine that only support row-based logging
General overview:
The logic for switching to row format when binlog_format=MIXED had
numerous flaws. The underlying problem was the lack of a consistent
architecture.
General purpose of this changeset:
This changeset introduces an architecture for switching to row format
when binlog_format=MIXED. It enforces the architecture where it has
to. It leaves some bugs to be fixed later. It adds extensive tests to
verify that unsafe statements work as expected and that appropriate
errors are produced by problems with the selection of binlog format.
It was not practical to split this into smaller pieces of work.
Problem 1:
To determine the logging mode, the code has to take several parameters
into account (namely: (1) the value of binlog_format; (2) the
capabilities of the engines; (3) the type of the current statement:
normal, unsafe, or row injection). These parameters may conflict in
several ways, namely:
- binlog_format=STATEMENT for a row injection
- binlog_format=STATEMENT for an unsafe statement
- binlog_format=STATEMENT for an engine only supporting row logging
- binlog_format=ROW for an engine only supporting statement logging
- statement is unsafe and engine does not support row logging
- row injection in a table that does not support statement logging
- statement modifies one table that does not support row logging and
one that does not support statement logging
Several of these conflicts were not detected, or were detected with
an inappropriate error message. The problem of BUG#39934 was that no
appropriate error message was written for the case when an engine
only supporting row logging executed a row injection with
binlog_format=ROW. However, all above cases must be handled.
Fix 1:
Introduce new error codes (sql/share/errmsg.txt). Ensure that all
conditions are detected and handled in decide_logging_format()
Problem 2:
The binlog format shall be determined once per statement, in
decide_logging_format(). It shall not be changed before or after that.
Before decide_logging_format() is called, all information necessary to
determine the logging format must be available. This principle ensures
that all unsafe statements are handled in a consistent way.
However, this principle is not followed:
thd->set_current_stmt_binlog_row_based_if_mixed() is called in several
places, including from code executing UPDATE..LIMIT,
INSERT..SELECT..LIMIT, DELETE..LIMIT, INSERT DELAYED, and
SET @@binlog_format. After Problem 1 was fixed, that caused
inconsistencies where these unsafe statements would not print the
appropriate warnings or errors for some of the conflicts.
Fix 2:
Remove calls to THD::set_current_stmt_binlog_row_based_if_mixed() from
code executed after decide_logging_format(). Compensate by calling the
set_current_stmt_unsafe() at parse time. This way, all unsafe statements
are detected by decide_logging_format().
Problem 3:
INSERT DELAYED is not unsafe: it is logged in statement format even if
binlog_format=MIXED, and no warning is printed even if
binlog_format=STATEMENT. This is BUG#45825.
Fix 3:
Made INSERT DELAYED set itself to unsafe at parse time. This allows
decide_logging_format() to detect that a warning should be printed or
the binlog_format changed.
Problem 4:
LIMIT clause were not marked as unsafe when executed inside stored
functions/triggers/views/prepared statements. This is
BUG#45785.
Fix 4:
Make statements containing the LIMIT clause marked as unsafe at
parse time, instead of at execution time. This allows propagating
unsafe-ness to the view.
17 years ago BUG#39934: Slave stops for engine that only support row-based logging
This is a post-push fix addressing review requests and
problems with extra warnings.
Problem 1: The sub-statement where an unsafe warning was detected was
printed as part of the warning. This was ok for statements that
were unsafe due to, e.g., calls to UUID(), but did not make
sense for statements that were unsafe because there was more than
one autoincrement column (unsafeness in this case comes from the
combination of several sub-statements).
Fix 1: Instead of printing the sub-statement, print an explanation
of why the statement is unsafe.
Problem 2:
When a recursive construct (i.e., stored proceure, stored
function, trigger, view, prepared statement) contained several
sub-statements, and at least one of them was unsafe, there would be
one unsafeness warning per sub-statement - even for safe
sub-statements.
Fix 2:
Ensure that each type of warning is printed at most once, by
remembering throughout the execution of the statement which types
of warnings have been printed.
17 years ago BUG#39934: Slave stops for engine that only support row-based logging
General overview:
The logic for switching to row format when binlog_format=MIXED had
numerous flaws. The underlying problem was the lack of a consistent
architecture.
General purpose of this changeset:
This changeset introduces an architecture for switching to row format
when binlog_format=MIXED. It enforces the architecture where it has
to. It leaves some bugs to be fixed later. It adds extensive tests to
verify that unsafe statements work as expected and that appropriate
errors are produced by problems with the selection of binlog format.
It was not practical to split this into smaller pieces of work.
Problem 1:
To determine the logging mode, the code has to take several parameters
into account (namely: (1) the value of binlog_format; (2) the
capabilities of the engines; (3) the type of the current statement:
normal, unsafe, or row injection). These parameters may conflict in
several ways, namely:
- binlog_format=STATEMENT for a row injection
- binlog_format=STATEMENT for an unsafe statement
- binlog_format=STATEMENT for an engine only supporting row logging
- binlog_format=ROW for an engine only supporting statement logging
- statement is unsafe and engine does not support row logging
- row injection in a table that does not support statement logging
- statement modifies one table that does not support row logging and
one that does not support statement logging
Several of these conflicts were not detected, or were detected with
an inappropriate error message. The problem of BUG#39934 was that no
appropriate error message was written for the case when an engine
only supporting row logging executed a row injection with
binlog_format=ROW. However, all above cases must be handled.
Fix 1:
Introduce new error codes (sql/share/errmsg.txt). Ensure that all
conditions are detected and handled in decide_logging_format()
Problem 2:
The binlog format shall be determined once per statement, in
decide_logging_format(). It shall not be changed before or after that.
Before decide_logging_format() is called, all information necessary to
determine the logging format must be available. This principle ensures
that all unsafe statements are handled in a consistent way.
However, this principle is not followed:
thd->set_current_stmt_binlog_row_based_if_mixed() is called in several
places, including from code executing UPDATE..LIMIT,
INSERT..SELECT..LIMIT, DELETE..LIMIT, INSERT DELAYED, and
SET @@binlog_format. After Problem 1 was fixed, that caused
inconsistencies where these unsafe statements would not print the
appropriate warnings or errors for some of the conflicts.
Fix 2:
Remove calls to THD::set_current_stmt_binlog_row_based_if_mixed() from
code executed after decide_logging_format(). Compensate by calling the
set_current_stmt_unsafe() at parse time. This way, all unsafe statements
are detected by decide_logging_format().
Problem 3:
INSERT DELAYED is not unsafe: it is logged in statement format even if
binlog_format=MIXED, and no warning is printed even if
binlog_format=STATEMENT. This is BUG#45825.
Fix 3:
Made INSERT DELAYED set itself to unsafe at parse time. This allows
decide_logging_format() to detect that a warning should be printed or
the binlog_format changed.
Problem 4:
LIMIT clause were not marked as unsafe when executed inside stored
functions/triggers/views/prepared statements. This is
BUG#45785.
Fix 4:
Make statements containing the LIMIT clause marked as unsafe at
parse time, instead of at execution time. This allows propagating
unsafe-ness to the view.
17 years ago Bug#34043: Server loops excessively in _checkchunk() when safemalloc is enabled
Essentially, the problem is that safemalloc is excruciatingly
slow as it checks all allocated blocks for overrun at each
memory management primitive, yielding a almost exponential
slowdown for the memory management functions (malloc, realloc,
free). The overrun check basically consists of verifying some
bytes of a block for certain magic keys, which catches some
simple forms of overrun. Another minor problem is violation
of aliasing rules and that its own internal list of blocks
is prone to corruption.
Another issue with safemalloc is rather the maintenance cost
as the tool has a significant impact on the server code.
Given the magnitude of memory debuggers available nowadays,
especially those that are provided with the platform malloc
implementation, maintenance of a in-house and largely obsolete
memory debugger becomes a burden that is not worth the effort
due to its slowness and lack of support for detecting more
common forms of heap corruption.
Since there are third-party tools that can provide the same
functionality at a lower or comparable performance cost, the
solution is to simply remove safemalloc. Third-party tools
can provide the same functionality at a lower or comparable
performance cost.
The removal of safemalloc also allows a simplification of the
malloc wrappers, removing quite a bit of kludge: redefinition
of my_malloc, my_free and the removal of the unused second
argument of my_free. Since free() always check whether the
supplied pointer is null, redudant checks are also removed.
Also, this patch adds unit testing for my_malloc and moves
my_realloc implementation into the same file as the other
memory allocation primitives. 16 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#19022 "Memory bug when switching db during trigger execution"
Bug#17199 "Problem when view calls function from another database."
Bug#18444 "Fully qualified stored function names don't work correctly in
SELECT statements"
Documentation note: this patch introduces a change in behaviour of prepared
statements.
This patch adds a few new invariants with regard to how THD::db should
be used. These invariants should be preserved in future:
- one should never refer to THD::db by pointer and always make a deep copy
(strmake, strdup)
- one should never compare two databases by pointer, but use strncmp or
my_strncasecmp
- TABLE_LIST object table->db should be always initialized in the parser or
by creator of the object.
For prepared statements it means that if the current database is changed
after a statement is prepared, the database that was current at prepare
remains active. This also means that you can not prepare a statement that
implicitly refers to the current database if the latter is not set.
This is not documented, and therefore needs documentation. This is NOT a
change in behavior for almost all SQL statements except:
- ALTER TABLE t1 RENAME t2
- OPTIMIZE TABLE t1
- ANALYZE TABLE t1
- TRUNCATE TABLE t1 --
until this patch t1 or t2 could be evaluated at the first execution of
prepared statement.
CURRENT_DATABASE() still works OK and is evaluated at every execution
of prepared statement.
Note, that in stored routines this is not an issue as the default
database is the database of the stored procedure and "use" statement
is prohibited in stored routines.
This patch makes obsolete the use of check_db_used (it was never used in the
old code too) and all other places that check for table->db and assign it
from THD::db if it's NULL, except the parser.
How this patch was created: THD::{db,db_length} were replaced with a
LEX_STRING, THD::db. All the places that refer to THD::{db,db_length} were
manually checked and:
- if the place uses thd->db by pointer, it was fixed to make a deep copy
- if a place compared two db pointers, it was fixed to compare them by value
(via strcmp/my_strcasecmp, whatever was approproate)
Then this intermediate patch was used to write a smaller patch that does the
same thing but without a rename.
TODO in 5.1:
- remove check_db_used
- deploy THD::set_db in mysql_change_db
See also comments to individual files.
20 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
19 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
19 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
19 years ago Bug#34043: Server loops excessively in _checkchunk() when safemalloc is enabled
Essentially, the problem is that safemalloc is excruciatingly
slow as it checks all allocated blocks for overrun at each
memory management primitive, yielding a almost exponential
slowdown for the memory management functions (malloc, realloc,
free). The overrun check basically consists of verifying some
bytes of a block for certain magic keys, which catches some
simple forms of overrun. Another minor problem is violation
of aliasing rules and that its own internal list of blocks
is prone to corruption.
Another issue with safemalloc is rather the maintenance cost
as the tool has a significant impact on the server code.
Given the magnitude of memory debuggers available nowadays,
especially those that are provided with the platform malloc
implementation, maintenance of a in-house and largely obsolete
memory debugger becomes a burden that is not worth the effort
due to its slowness and lack of support for detecting more
common forms of heap corruption.
Since there are third-party tools that can provide the same
functionality at a lower or comparable performance cost, the
solution is to simply remove safemalloc. Third-party tools
can provide the same functionality at a lower or comparable
performance cost.
The removal of safemalloc also allows a simplification of the
malloc wrappers, removing quite a bit of kludge: redefinition
of my_malloc, my_free and the removal of the unused second
argument of my_free. Since free() always check whether the
supplied pointer is null, redudant checks are also removed.
Also, this patch adds unit testing for my_malloc and moves
my_realloc implementation into the same file as the other
memory allocation primitives. 16 years ago WL#3146 "less locking in auto_increment":
this is a cleanup patch for our current auto_increment handling:
new names for auto_increment variables in THD, new methods to manipulate them
(see sql_class.h), some move into handler::, causing less backup/restore
work when executing substatements.
This makes the logic hopefully clearer, less work is is needed in
mysql_insert().
By cleaning up, using different variables for different purposes (instead
of one for 3 things...), we fix those bugs, which someone may want to fix
in 5.0 too:
BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate
statement-based"
BUG#20341 "stored function inserting into one auto_increment puts bad
data in slave"
BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE"
(now if a row is updated, LAST_INSERT_ID() will return its id)
and re-fixes:
BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT"
(already fixed differently by Ramil in 4.1)
Test of documented behaviour of mysql_insert_id() (there was no test).
The behaviour changes introduced are:
- LAST_INSERT_ID() now returns "the first autogenerated auto_increment value
successfully inserted", instead of "the first autogenerated auto_increment
value if any row was successfully inserted", see auto_increment.test.
Same for mysql_insert_id(), see mysql_client_test.c.
- LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY
UPDATE, see auto_increment.test. Same for mysql_insert_id(), see
mysql_client_test.c.
- LAST_INSERT_ID() does not change if no autogenerated value was successfully
inserted (it used to then be 0), see auto_increment.test.
- if in INSERT SELECT no autogenerated value was successfully inserted,
mysql_insert_id() now returns the id of the last inserted row (it already
did this for INSERT VALUES), see mysql_client_test.c.
- if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X
(it already did this for INSERT VALUES), see mysql_client_test.c.
- NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE,
the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID
influences not only the first row now.
Additionally, when unlocking a table we check that the thread is not keeping
a next_insert_id (as the table is unlocked that id is potentially out-of-date);
forgetting about this next_insert_id is done in a new
handler::ha_release_auto_increment().
Finally we prepare for engines capable of reserving finite-length intervals
of auto_increment values: we store such intervals in THD. The next step
(to be done by the replication team in 5.1) is to read those intervals from
THD and actually store them in the statement-based binary log. NDB
will be a good engine to test that.
20 years ago A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
19 years ago WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
19 years ago Bug#34043: Server loops excessively in _checkchunk() when safemalloc is enabled
Essentially, the problem is that safemalloc is excruciatingly
slow as it checks all allocated blocks for overrun at each
memory management primitive, yielding a almost exponential
slowdown for the memory management functions (malloc, realloc,
free). The overrun check basically consists of verifying some
bytes of a block for certain magic keys, which catches some
simple forms of overrun. Another minor problem is violation
of aliasing rules and that its own internal list of blocks
is prone to corruption.
Another issue with safemalloc is rather the maintenance cost
as the tool has a significant impact on the server code.
Given the magnitude of memory debuggers available nowadays,
especially those that are provided with the platform malloc
implementation, maintenance of a in-house and largely obsolete
memory debugger becomes a burden that is not worth the effort
due to its slowness and lack of support for detecting more
common forms of heap corruption.
Since there are third-party tools that can provide the same
functionality at a lower or comparable performance cost, the
solution is to simply remove safemalloc. Third-party tools
can provide the same functionality at a lower or comparable
performance cost.
The removal of safemalloc also allows a simplification of the
malloc wrappers, removing quite a bit of kludge: redefinition
of my_malloc, my_free and the removal of the unused second
argument of my_free. Since free() always check whether the
supplied pointer is null, redudant checks are also removed.
Also, this patch adds unit testing for my_malloc and moves
my_realloc implementation into the same file as the other
memory allocation primitives. 16 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
19 years ago WL#3146 "less locking in auto_increment":
this is a cleanup patch for our current auto_increment handling:
new names for auto_increment variables in THD, new methods to manipulate them
(see sql_class.h), some move into handler::, causing less backup/restore
work when executing substatements.
This makes the logic hopefully clearer, less work is is needed in
mysql_insert().
By cleaning up, using different variables for different purposes (instead
of one for 3 things...), we fix those bugs, which someone may want to fix
in 5.0 too:
BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate
statement-based"
BUG#20341 "stored function inserting into one auto_increment puts bad
data in slave"
BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE"
(now if a row is updated, LAST_INSERT_ID() will return its id)
and re-fixes:
BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT"
(already fixed differently by Ramil in 4.1)
Test of documented behaviour of mysql_insert_id() (there was no test).
The behaviour changes introduced are:
- LAST_INSERT_ID() now returns "the first autogenerated auto_increment value
successfully inserted", instead of "the first autogenerated auto_increment
value if any row was successfully inserted", see auto_increment.test.
Same for mysql_insert_id(), see mysql_client_test.c.
- LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY
UPDATE, see auto_increment.test. Same for mysql_insert_id(), see
mysql_client_test.c.
- LAST_INSERT_ID() does not change if no autogenerated value was successfully
inserted (it used to then be 0), see auto_increment.test.
- if in INSERT SELECT no autogenerated value was successfully inserted,
mysql_insert_id() now returns the id of the last inserted row (it already
did this for INSERT VALUES), see mysql_client_test.c.
- if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X
(it already did this for INSERT VALUES), see mysql_client_test.c.
- NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE,
the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID
influences not only the first row now.
Additionally, when unlocking a table we check that the thread is not keeping
a next_insert_id (as the table is unlocked that id is potentially out-of-date);
forgetting about this next_insert_id is done in a new
handler::ha_release_auto_increment().
Finally we prepare for engines capable of reserving finite-length intervals
of auto_increment values: we store such intervals in THD. The next step
(to be done by the replication team in 5.1) is to read those intervals from
THD and actually store them in the statement-based binary log. NDB
will be a good engine to test that.
20 years ago WL#3146 "less locking in auto_increment":
this is a cleanup patch for our current auto_increment handling:
new names for auto_increment variables in THD, new methods to manipulate them
(see sql_class.h), some move into handler::, causing less backup/restore
work when executing substatements.
This makes the logic hopefully clearer, less work is is needed in
mysql_insert().
By cleaning up, using different variables for different purposes (instead
of one for 3 things...), we fix those bugs, which someone may want to fix
in 5.0 too:
BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate
statement-based"
BUG#20341 "stored function inserting into one auto_increment puts bad
data in slave"
BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE"
(now if a row is updated, LAST_INSERT_ID() will return its id)
and re-fixes:
BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT"
(already fixed differently by Ramil in 4.1)
Test of documented behaviour of mysql_insert_id() (there was no test).
The behaviour changes introduced are:
- LAST_INSERT_ID() now returns "the first autogenerated auto_increment value
successfully inserted", instead of "the first autogenerated auto_increment
value if any row was successfully inserted", see auto_increment.test.
Same for mysql_insert_id(), see mysql_client_test.c.
- LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY
UPDATE, see auto_increment.test. Same for mysql_insert_id(), see
mysql_client_test.c.
- LAST_INSERT_ID() does not change if no autogenerated value was successfully
inserted (it used to then be 0), see auto_increment.test.
- if in INSERT SELECT no autogenerated value was successfully inserted,
mysql_insert_id() now returns the id of the last inserted row (it already
did this for INSERT VALUES), see mysql_client_test.c.
- if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X
(it already did this for INSERT VALUES), see mysql_client_test.c.
- NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE,
the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID
influences not only the first row now.
Additionally, when unlocking a table we check that the thread is not keeping
a next_insert_id (as the table is unlocked that id is potentially out-of-date);
forgetting about this next_insert_id is done in a new
handler::ha_release_auto_increment().
Finally we prepare for engines capable of reserving finite-length intervals
of auto_increment values: we store such intervals in THD. The next step
(to be done by the replication team in 5.1) is to read those intervals from
THD and actually store them in the statement-based binary log. NDB
will be a good engine to test that.
20 years ago WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
19 years ago Manual merge from 5.0-rpl, of fixes for:
1)
BUG#25507 "multi-row insert delayed + auto increment causes
duplicate key entries on slave" (two concurrrent connections doing
multi-row INSERT DELAYED to insert into an auto_increment column,
caused replication slave to stop with "duplicate key error" (and
binlog was wrong), and BUG#26116 "If multi-row INSERT
DELAYED has errors, statement-based binlogging breaks" (the binlog
was not accounting for all rows inserted, or slave could stop).
The fix is that: in statement-based binlogging, a multi-row INSERT
DELAYED is silently converted to a non-delayed INSERT.
This is supposed to not affect many 5.1 users as in 5.1, the default
binlog format is "mixed", which does not have the bug (the bug is
only with binlog_format=STATEMENT).
We should document how the system delayed_insert thread decides of
its binlog format (which is not modified by this patch):
this decision is taken when the thread is created
and holds until it is terminated (is not affected by any later change
via SET GLOBAL BINLOG_FORMAT). It is also not affected by the binlog
format of the connection which issues INSERT DELAYED (this binlog
format does not affect how the row will be binlogged).
If one wants to change the binlog format of its server with SET
GLOBAL BINLOG_FORMAT, it should do FLUSH TABLES to be sure all
delayed_insert threads terminate and thus new threads are created,
taking into account the new format.
2)
BUG#24432
"INSERT... ON DUPLICATE KEY UPDATE skips auto_increment values".
When in an INSERT ON DUPLICATE KEY UPDATE, using
an autoincrement column, we inserted some autogenerated values and
also updated some rows, some autogenerated values were not used
(for example, even if 10 was the largest autoinc value in the table
at the start of the statement, 12 could be the first autogenerated
value inserted by the statement, instead of 11). One autogenerated
value was lost per updated row. Led to exhausting the range of the
autoincrement column faster.
Bug introduced by fix of BUG#20188; present since 5.0.24 and 5.1.12.
This bug breaks replication from a pre-5.0.24/pre-5.1.12 master.
But the present bugfix, as it makes INSERT ON DUP KEY UPDATE
behave like pre-5.0.24/pre-5.1.12, breaks replication from a
[5.0.24,5.0.34]/[5.1.12,5.1.15]
master to a fixed (5.0.36/5.1.16) slave! To warn users against this when
they upgrade their slave, as agreed with the support team, we add
code for a fixed slave to detect that it is connected to a buggy
master in a situation (INSERT ON DUP KEY UPDATE into autoinc column)
likely to break replication, in which case it cannot replicate so
stops and prints a message to the slave's error log and to SHOW SLAVE
STATUS.
For 5.0.36->[5.0.24,5.0.34] replication or 5.1.16->[5.1.12,5.1.15]
replication we cannot warn as master
does not know the slave's version (but we always recommended to users
to have slave at least as new as master).
As agreed with support, I have asked for an alert to be put into
the MySQL Network Monitoring and Advisory Service.
3) note that I'll re-enable rpl_insert_id as soon as 5.1-rpl gets
the changes from the main 5.1.
19 years ago BUG#37426: RBR breaks for CHAR() UTF-8 fields > 85 chars
In order to handle CHAR() fields, 8 bits were reserved for
the size of the CHAR field. However, instead of denoting the
number of characters in the field, field_length was used which
denotes the number of bytes in the field.
Since UTF-8 fields can have three bytes per character (and
has been extended to have four bytes per character in 6.0),
an extra two bits have been encoded in the field metadata
work for fields of type Field_string (i.e., CHAR fields).
Since the metadata word is filled, the extra bits have been
encoded in the upper 4 bits of the real type (the most
significant byte of the metadata word) by computing the
bitwise xor of the extra two bits. Since the upper 4 bits
of the real type always is 1111 for Field_string, this
means that for fields of length <256, the encoding is
identical to the encoding used in pre-5.1.26 servers, but
for lengths of 256 or more, an unrecognized type is formed,
causing an old slave (that does not handle lengths of 256
or more) to stop.
18 years ago Manual merge from 5.0-rpl, of fixes for:
1)
BUG#25507 "multi-row insert delayed + auto increment causes
duplicate key entries on slave" (two concurrrent connections doing
multi-row INSERT DELAYED to insert into an auto_increment column,
caused replication slave to stop with "duplicate key error" (and
binlog was wrong), and BUG#26116 "If multi-row INSERT
DELAYED has errors, statement-based binlogging breaks" (the binlog
was not accounting for all rows inserted, or slave could stop).
The fix is that: in statement-based binlogging, a multi-row INSERT
DELAYED is silently converted to a non-delayed INSERT.
This is supposed to not affect many 5.1 users as in 5.1, the default
binlog format is "mixed", which does not have the bug (the bug is
only with binlog_format=STATEMENT).
We should document how the system delayed_insert thread decides of
its binlog format (which is not modified by this patch):
this decision is taken when the thread is created
and holds until it is terminated (is not affected by any later change
via SET GLOBAL BINLOG_FORMAT). It is also not affected by the binlog
format of the connection which issues INSERT DELAYED (this binlog
format does not affect how the row will be binlogged).
If one wants to change the binlog format of its server with SET
GLOBAL BINLOG_FORMAT, it should do FLUSH TABLES to be sure all
delayed_insert threads terminate and thus new threads are created,
taking into account the new format.
2)
BUG#24432
"INSERT... ON DUPLICATE KEY UPDATE skips auto_increment values".
When in an INSERT ON DUPLICATE KEY UPDATE, using
an autoincrement column, we inserted some autogenerated values and
also updated some rows, some autogenerated values were not used
(for example, even if 10 was the largest autoinc value in the table
at the start of the statement, 12 could be the first autogenerated
value inserted by the statement, instead of 11). One autogenerated
value was lost per updated row. Led to exhausting the range of the
autoincrement column faster.
Bug introduced by fix of BUG#20188; present since 5.0.24 and 5.1.12.
This bug breaks replication from a pre-5.0.24/pre-5.1.12 master.
But the present bugfix, as it makes INSERT ON DUP KEY UPDATE
behave like pre-5.0.24/pre-5.1.12, breaks replication from a
[5.0.24,5.0.34]/[5.1.12,5.1.15]
master to a fixed (5.0.36/5.1.16) slave! To warn users against this when
they upgrade their slave, as agreed with the support team, we add
code for a fixed slave to detect that it is connected to a buggy
master in a situation (INSERT ON DUP KEY UPDATE into autoinc column)
likely to break replication, in which case it cannot replicate so
stops and prints a message to the slave's error log and to SHOW SLAVE
STATUS.
For 5.0.36->[5.0.24,5.0.34] replication or 5.1.16->[5.1.12,5.1.15]
replication we cannot warn as master
does not know the slave's version (but we always recommended to users
to have slave at least as new as master).
As agreed with support, I have asked for an alert to be put into
the MySQL Network Monitoring and Advisory Service.
3) note that I'll re-enable rpl_insert_id as soon as 5.1-rpl gets
the changes from the main 5.1.
19 years ago WL#3146 "less locking in auto_increment":
this is a cleanup patch for our current auto_increment handling:
new names for auto_increment variables in THD, new methods to manipulate them
(see sql_class.h), some move into handler::, causing less backup/restore
work when executing substatements.
This makes the logic hopefully clearer, less work is is needed in
mysql_insert().
By cleaning up, using different variables for different purposes (instead
of one for 3 things...), we fix those bugs, which someone may want to fix
in 5.0 too:
BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate
statement-based"
BUG#20341 "stored function inserting into one auto_increment puts bad
data in slave"
BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE"
(now if a row is updated, LAST_INSERT_ID() will return its id)
and re-fixes:
BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT"
(already fixed differently by Ramil in 4.1)
Test of documented behaviour of mysql_insert_id() (there was no test).
The behaviour changes introduced are:
- LAST_INSERT_ID() now returns "the first autogenerated auto_increment value
successfully inserted", instead of "the first autogenerated auto_increment
value if any row was successfully inserted", see auto_increment.test.
Same for mysql_insert_id(), see mysql_client_test.c.
- LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY
UPDATE, see auto_increment.test. Same for mysql_insert_id(), see
mysql_client_test.c.
- LAST_INSERT_ID() does not change if no autogenerated value was successfully
inserted (it used to then be 0), see auto_increment.test.
- if in INSERT SELECT no autogenerated value was successfully inserted,
mysql_insert_id() now returns the id of the last inserted row (it already
did this for INSERT VALUES), see mysql_client_test.c.
- if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X
(it already did this for INSERT VALUES), see mysql_client_test.c.
- NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE,
the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID
influences not only the first row now.
Additionally, when unlocking a table we check that the thread is not keeping
a next_insert_id (as the table is unlocked that id is potentially out-of-date);
forgetting about this next_insert_id is done in a new
handler::ha_release_auto_increment().
Finally we prepare for engines capable of reserving finite-length intervals
of auto_increment values: we store such intervals in THD. The next step
(to be done by the replication team in 5.1) is to read those intervals from
THD and actually store them in the statement-based binary log. NDB
will be a good engine to test that.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago WL#3146 "less locking in auto_increment":
this is a cleanup patch for our current auto_increment handling:
new names for auto_increment variables in THD, new methods to manipulate them
(see sql_class.h), some move into handler::, causing less backup/restore
work when executing substatements.
This makes the logic hopefully clearer, less work is is needed in
mysql_insert().
By cleaning up, using different variables for different purposes (instead
of one for 3 things...), we fix those bugs, which someone may want to fix
in 5.0 too:
BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate
statement-based"
BUG#20341 "stored function inserting into one auto_increment puts bad
data in slave"
BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE"
(now if a row is updated, LAST_INSERT_ID() will return its id)
and re-fixes:
BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT"
(already fixed differently by Ramil in 4.1)
Test of documented behaviour of mysql_insert_id() (there was no test).
The behaviour changes introduced are:
- LAST_INSERT_ID() now returns "the first autogenerated auto_increment value
successfully inserted", instead of "the first autogenerated auto_increment
value if any row was successfully inserted", see auto_increment.test.
Same for mysql_insert_id(), see mysql_client_test.c.
- LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY
UPDATE, see auto_increment.test. Same for mysql_insert_id(), see
mysql_client_test.c.
- LAST_INSERT_ID() does not change if no autogenerated value was successfully
inserted (it used to then be 0), see auto_increment.test.
- if in INSERT SELECT no autogenerated value was successfully inserted,
mysql_insert_id() now returns the id of the last inserted row (it already
did this for INSERT VALUES), see mysql_client_test.c.
- if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X
(it already did this for INSERT VALUES), see mysql_client_test.c.
- NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE,
the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID
influences not only the first row now.
Additionally, when unlocking a table we check that the thread is not keeping
a next_insert_id (as the table is unlocked that id is potentially out-of-date);
forgetting about this next_insert_id is done in a new
handler::ha_release_auto_increment().
Finally we prepare for engines capable of reserving finite-length intervals
of auto_increment values: we store such intervals in THD. The next step
(to be done by the replication team in 5.1) is to read those intervals from
THD and actually store them in the statement-based binary log. NDB
will be a good engine to test that.
20 years ago 5.1 version of a fix and test cases for bugs:
Bug#4968 ""Stored procedure crash if cursor opened on altered table"
Bug#6895 "Prepared Statements: ALTER TABLE DROP COLUMN does nothing"
Bug#19182 "CREATE TABLE bar (m INT) SELECT n FROM foo; doesn't work from
stored procedure."
Bug#19733 "Repeated alter, or repeated create/drop, fails"
Bug#22060 "ALTER TABLE x AUTO_INCREMENT=y in SP crashes server"
Bug#24879 "Prepared Statements: CREATE TABLE (UTF8 KEY) produces a
growing key length" (this bug is not fixed in 5.0)
Re-execution of CREATE DATABASE, CREATE TABLE and ALTER TABLE
statements in stored routines or as prepared statements caused
incorrect results (and crashes in versions prior to 5.0.25).
In 5.1 the problem occured only for CREATE DATABASE, CREATE TABLE
SELECT and CREATE TABLE with INDEX/DATA DIRECTOY options).
The problem of bugs 4968, 19733, 19282 and 6895 was that functions
mysql_prepare_table, mysql_create_table and mysql_alter_table are not
re-execution friendly: during their operation they modify contents
of LEX (members create_info, alter_info, key_list, create_list),
thus making the LEX unusable for the next execution.
In particular, these functions removed processed columns and keys from
create_list, key_list and drop_list. Search the code in sql_table.cc
for drop_it.remove() and similar patterns to find evidence.
The fix is to supply to these functions a usable copy of each of the
above structures at every re-execution of an SQL statement.
To simplify memory management, LEX::key_list and LEX::create_list
were added to LEX::alter_info, a fresh copy of which is created for
every execution.
The problem of crashing bug 22060 stemmed from the fact that the above
metnioned functions were not only modifying HA_CREATE_INFO structure
in LEX, but also were changing it to point to areas in volatile memory
of the execution memory root.
The patch solves this problem by creating and using an on-stack
copy of HA_CREATE_INFO in mysql_execute_command.
Additionally, this patch splits the part of mysql_alter_table
that analizes and rewrites information from the parser into
a separate function - mysql_prepare_alter_table, in analogy with
mysql_prepare_table, which is renamed to mysql_prepare_create_table.
19 years ago Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE
corrupts a MERGE table
Bug 26867 - LOCK TABLES + REPAIR + merge table result in
memory/cpu hogging
Bug 26377 - Deadlock with MERGE and FLUSH TABLE
Bug 25038 - Waiting TRUNCATE
Bug 25700 - merge base tables get corrupted by
optimize/analyze/repair table
Bug 30275 - Merge tables: flush tables or unlock tables
causes server to crash
Bug 19627 - temporary merge table locking
Bug 27660 - Falcon: merge table possible
Bug 30273 - merge tables: Can't lock file (errno: 155)
The problems were:
Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE
corrupts a MERGE table
1. A thread trying to lock a MERGE table performs busy waiting while
REPAIR TABLE or a similar table administration task is ongoing on
one or more of its MyISAM tables.
2. A thread trying to lock a MERGE table performs busy waiting until all
threads that did REPAIR TABLE or similar table administration tasks
on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK
TABLES. The difference against problem #1 is that the busy waiting
takes place *after* the administration task. It is terminated by
UNLOCK TABLES only.
3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the
lock. This does *not* require a MERGE table. The first FLUSH TABLES
can be replaced by any statement that requires other threads to
reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke
the problem.
Bug 26867 - LOCK TABLES + REPAIR + merge table result in
memory/cpu hogging
Trying DML on a MERGE table, which has a child locked and
repaired by another thread, made an infinite loop in the server.
Bug 26377 - Deadlock with MERGE and FLUSH TABLE
Locking a MERGE table and its children in parent-child order
and flushing the child deadlocked the server.
Bug 25038 - Waiting TRUNCATE
Truncating a MERGE child, while the MERGE table was in use,
let the truncate fail instead of waiting for the table to
become free.
Bug 25700 - merge base tables get corrupted by
optimize/analyze/repair table
Repairing a child of an open MERGE table corrupted the child.
It was necessary to FLUSH the child first.
Bug 30275 - Merge tables: flush tables or unlock tables
causes server to crash
Flushing and optimizing locked MERGE children crashed the server.
Bug 19627 - temporary merge table locking
Use of a temporary MERGE table with non-temporary children
could corrupt the children.
Temporary tables are never locked. So we do now prohibit
non-temporary chidlren of a temporary MERGE table.
Bug 27660 - Falcon: merge table possible
It was possible to create a MERGE table with non-MyISAM children.
Bug 30273 - merge tables: Can't lock file (errno: 155)
This was a Windows-only bug. Table administration statements
sometimes failed with "Can't lock file (errno: 155)".
These bugs are fixed by a new implementation of MERGE table open.
When opening a MERGE table in open_tables() we do now add the
child tables to the list of tables to be opened by open_tables()
(the "query_list"). The children are not opened in the handler at
this stage.
After opening the parent, open_tables() opens each child from the
now extended query_list. When the last child is opened, we remove
the children from the query_list again and attach the children to
the parent. This behaves similar to the old open. However it does
not open the MyISAM tables directly, but grabs them from the already
open children.
When closing a MERGE table in close_thread_table() we detach the
children only. Closing of the children is done implicitly because
they are in thd->open_tables.
For more detail see the comment at the top of ha_myisammrg.cc.
Changed from open_ltable() to open_and_lock_tables() in all places
that can be relevant for MERGE tables. The latter can handle tables
added to the list on the fly. When open_ltable() was used in a loop
over a list of tables, the list must be temporarily terminated
after every table for open_and_lock_tables().
table_list->required_type is set to FRMTYPE_TABLE to avoid open of
special tables. Handling of derived tables is suppressed.
These details are handled by the new function
open_n_lock_single_table(), which has nearly the same signature as
open_ltable() and can replace it in most cases.
In reopen_tables() some of the tables open by a thread can be
closed and reopened. When a MERGE child is affected, the parent
must be closed and reopened too. Closing of the parent is forced
before the first child is closed. Reopen happens in the order of
thd->open_tables. MERGE parents do not attach their children
automatically at open. This is done after all tables are reopened.
So all children are open when attaching them.
Special lock handling like mysql_lock_abort() or mysql_lock_remove()
needs to be suppressed for MERGE children or forwarded to the parent.
This depends on the situation. In loops over all open tables one
suppresses child lock handling. When a single table is touched,
forwarding is done.
Behavioral changes:
===================
This patch changes the behavior of temporary MERGE tables.
Temporary MERGE must have temporary children.
The old behavior was wrong. A temporary table is not locked. Hence
even non-temporary children were not locked. See
Bug 19627 - temporary merge table locking.
You cannot change the union list of a non-temporary MERGE table
when LOCK TABLES is in effect. The following does *not* work:
CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...;
LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE;
ALTER TABLE m1 ... UNION=(t1,t2) ...;
However, you can do this with a temporary MERGE table.
You cannot create a MERGE table with CREATE ... SELECT, neither
as a temporary MERGE table, nor as a non-temporary MERGE table.
CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...;
Gives error message: table is not BASE TABLE.
18 years ago 5.1 version of a fix and test cases for bugs:
Bug#4968 ""Stored procedure crash if cursor opened on altered table"
Bug#6895 "Prepared Statements: ALTER TABLE DROP COLUMN does nothing"
Bug#19182 "CREATE TABLE bar (m INT) SELECT n FROM foo; doesn't work from
stored procedure."
Bug#19733 "Repeated alter, or repeated create/drop, fails"
Bug#22060 "ALTER TABLE x AUTO_INCREMENT=y in SP crashes server"
Bug#24879 "Prepared Statements: CREATE TABLE (UTF8 KEY) produces a
growing key length" (this bug is not fixed in 5.0)
Re-execution of CREATE DATABASE, CREATE TABLE and ALTER TABLE
statements in stored routines or as prepared statements caused
incorrect results (and crashes in versions prior to 5.0.25).
In 5.1 the problem occured only for CREATE DATABASE, CREATE TABLE
SELECT and CREATE TABLE with INDEX/DATA DIRECTOY options).
The problem of bugs 4968, 19733, 19282 and 6895 was that functions
mysql_prepare_table, mysql_create_table and mysql_alter_table are not
re-execution friendly: during their operation they modify contents
of LEX (members create_info, alter_info, key_list, create_list),
thus making the LEX unusable for the next execution.
In particular, these functions removed processed columns and keys from
create_list, key_list and drop_list. Search the code in sql_table.cc
for drop_it.remove() and similar patterns to find evidence.
The fix is to supply to these functions a usable copy of each of the
above structures at every re-execution of an SQL statement.
To simplify memory management, LEX::key_list and LEX::create_list
were added to LEX::alter_info, a fresh copy of which is created for
every execution.
The problem of crashing bug 22060 stemmed from the fact that the above
metnioned functions were not only modifying HA_CREATE_INFO structure
in LEX, but also were changing it to point to areas in volatile memory
of the execution memory root.
The patch solves this problem by creating and using an on-stack
copy of HA_CREATE_INFO in mysql_execute_command.
Additionally, this patch splits the part of mysql_alter_table
that analizes and rewrites information from the parser into
a separate function - mysql_prepare_alter_table, in analogy with
mysql_prepare_table, which is renamed to mysql_prepare_create_table.
19 years ago 5.1 version of a fix and test cases for bugs:
Bug#4968 ""Stored procedure crash if cursor opened on altered table"
Bug#6895 "Prepared Statements: ALTER TABLE DROP COLUMN does nothing"
Bug#19182 "CREATE TABLE bar (m INT) SELECT n FROM foo; doesn't work from
stored procedure."
Bug#19733 "Repeated alter, or repeated create/drop, fails"
Bug#22060 "ALTER TABLE x AUTO_INCREMENT=y in SP crashes server"
Bug#24879 "Prepared Statements: CREATE TABLE (UTF8 KEY) produces a
growing key length" (this bug is not fixed in 5.0)
Re-execution of CREATE DATABASE, CREATE TABLE and ALTER TABLE
statements in stored routines or as prepared statements caused
incorrect results (and crashes in versions prior to 5.0.25).
In 5.1 the problem occured only for CREATE DATABASE, CREATE TABLE
SELECT and CREATE TABLE with INDEX/DATA DIRECTOY options).
The problem of bugs 4968, 19733, 19282 and 6895 was that functions
mysql_prepare_table, mysql_create_table and mysql_alter_table are not
re-execution friendly: during their operation they modify contents
of LEX (members create_info, alter_info, key_list, create_list),
thus making the LEX unusable for the next execution.
In particular, these functions removed processed columns and keys from
create_list, key_list and drop_list. Search the code in sql_table.cc
for drop_it.remove() and similar patterns to find evidence.
The fix is to supply to these functions a usable copy of each of the
above structures at every re-execution of an SQL statement.
To simplify memory management, LEX::key_list and LEX::create_list
were added to LEX::alter_info, a fresh copy of which is created for
every execution.
The problem of crashing bug 22060 stemmed from the fact that the above
metnioned functions were not only modifying HA_CREATE_INFO structure
in LEX, but also were changing it to point to areas in volatile memory
of the execution memory root.
The patch solves this problem by creating and using an on-stack
copy of HA_CREATE_INFO in mysql_execute_command.
Additionally, this patch splits the part of mysql_alter_table
that analizes and rewrites information from the parser into
a separate function - mysql_prepare_alter_table, in analogy with
mysql_prepare_table, which is renamed to mysql_prepare_create_table.
19 years ago  A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
16 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago BUG#22864 (Rollback following CREATE... SELECT discards 'CREATE TABLE'
from log):
When row-based logging is used, the CREATE-SELECT is written as two
parts: as a CREATE TABLE statement and as the rows for the table. For
both transactional and non-transactional tables, the CREATE TABLE
statement was written to the transaction cache, as were the rows, and
on statement end, the entire transaction cache was written to the binary
log if the table was non-transactional. For transactional tables, the
events were kept in the transaction cache until end of transaction (or
statement that were not part of a transaction).
For the case when AUTOCOMMIT=0 and we are creating a transactional table
using a create select, we would then keep the CREATE TABLE statement and
the rows for the CREATE-SELECT, while executing the following statements.
On a rollback, the transaction cache would then be cleared, which would
also remove the CREATE TABLE statement. Hence no table would be created
on the slave, while there is an empty table on the master.
This relates to BUG#22865 where the table being created exists on the
master, but not on the slave during insertion of rows into the newly
created table. This occurs since the CREATE TABLE statement were still
in the transaction cache until the statement finished executing, and
possibly longer if the table was transactional.
This patch changes the behaviour of the CREATE-SELECT statement by
adding an implicit commit at the end of the statement when creating
non-temporary tables. Hence, non-temporary tables will be written to the
binary log on completion, and in the even of AUTOCOMMIT=0, a new
transaction will be started. Temporary tables do not commit an ongoing
transaction: neither as a pre- not a post-commit.
The events for both transactional and non-transactional tables are
saved in the transaction cache, and written to the binary log at end
of the statement.
19 years ago BUG#39934: Slave stops for engine that only support row-based logging
General overview:
The logic for switching to row format when binlog_format=MIXED had
numerous flaws. The underlying problem was the lack of a consistent
architecture.
General purpose of this changeset:
This changeset introduces an architecture for switching to row format
when binlog_format=MIXED. It enforces the architecture where it has
to. It leaves some bugs to be fixed later. It adds extensive tests to
verify that unsafe statements work as expected and that appropriate
errors are produced by problems with the selection of binlog format.
It was not practical to split this into smaller pieces of work.
Problem 1:
To determine the logging mode, the code has to take several parameters
into account (namely: (1) the value of binlog_format; (2) the
capabilities of the engines; (3) the type of the current statement:
normal, unsafe, or row injection). These parameters may conflict in
several ways, namely:
- binlog_format=STATEMENT for a row injection
- binlog_format=STATEMENT for an unsafe statement
- binlog_format=STATEMENT for an engine only supporting row logging
- binlog_format=ROW for an engine only supporting statement logging
- statement is unsafe and engine does not support row logging
- row injection in a table that does not support statement logging
- statement modifies one table that does not support row logging and
one that does not support statement logging
Several of these conflicts were not detected, or were detected with
an inappropriate error message. The problem of BUG#39934 was that no
appropriate error message was written for the case when an engine
only supporting row logging executed a row injection with
binlog_format=ROW. However, all above cases must be handled.
Fix 1:
Introduce new error codes (sql/share/errmsg.txt). Ensure that all
conditions are detected and handled in decide_logging_format()
Problem 2:
The binlog format shall be determined once per statement, in
decide_logging_format(). It shall not be changed before or after that.
Before decide_logging_format() is called, all information necessary to
determine the logging format must be available. This principle ensures
that all unsafe statements are handled in a consistent way.
However, this principle is not followed:
thd->set_current_stmt_binlog_row_based_if_mixed() is called in several
places, including from code executing UPDATE..LIMIT,
INSERT..SELECT..LIMIT, DELETE..LIMIT, INSERT DELAYED, and
SET @@binlog_format. After Problem 1 was fixed, that caused
inconsistencies where these unsafe statements would not print the
appropriate warnings or errors for some of the conflicts.
Fix 2:
Remove calls to THD::set_current_stmt_binlog_row_based_if_mixed() from
code executed after decide_logging_format(). Compensate by calling the
set_current_stmt_unsafe() at parse time. This way, all unsafe statements
are detected by decide_logging_format().
Problem 3:
INSERT DELAYED is not unsafe: it is logged in statement format even if
binlog_format=MIXED, and no warning is printed even if
binlog_format=STATEMENT. This is BUG#45825.
Fix 3:
Made INSERT DELAYED set itself to unsafe at parse time. This allows
decide_logging_format() to detect that a warning should be printed or
the binlog_format changed.
Problem 4:
LIMIT clause were not marked as unsafe when executed inside stored
functions/triggers/views/prepared statements. This is
BUG#45785.
Fix 4:
Make statements containing the LIMIT clause marked as unsafe at
parse time, instead of at execution time. This allows propagating
unsafe-ness to the view.
17 years ago BUG#22864 (Rollback following CREATE... SELECT discards 'CREATE TABLE'
from log):
When row-based logging is used, the CREATE-SELECT is written as two
parts: as a CREATE TABLE statement and as the rows for the table. For
both transactional and non-transactional tables, the CREATE TABLE
statement was written to the transaction cache, as were the rows, and
on statement end, the entire transaction cache was written to the binary
log if the table was non-transactional. For transactional tables, the
events were kept in the transaction cache until end of transaction (or
statement that were not part of a transaction).
For the case when AUTOCOMMIT=0 and we are creating a transactional table
using a create select, we would then keep the CREATE TABLE statement and
the rows for the CREATE-SELECT, while executing the following statements.
On a rollback, the transaction cache would then be cleared, which would
also remove the CREATE TABLE statement. Hence no table would be created
on the slave, while there is an empty table on the master.
This relates to BUG#22865 where the table being created exists on the
master, but not on the slave during insertion of rows into the newly
created table. This occurs since the CREATE TABLE statement were still
in the transaction cache until the statement finished executing, and
possibly longer if the table was transactional.
This patch changes the behaviour of the CREATE-SELECT statement by
adding an implicit commit at the end of the statement when creating
non-temporary tables. Hence, non-temporary tables will be written to the
binary log on completion, and in the even of AUTOCOMMIT=0, a new
transaction will be started. Temporary tables do not commit an ongoing
transaction: neither as a pre- not a post-commit.
The events for both transactional and non-transactional tables are
saved in the transaction cache, and written to the binary log at end
of the statement.
19 years ago BUG#39934: Slave stops for engine that only support row-based logging
General overview:
The logic for switching to row format when binlog_format=MIXED had
numerous flaws. The underlying problem was the lack of a consistent
architecture.
General purpose of this changeset:
This changeset introduces an architecture for switching to row format
when binlog_format=MIXED. It enforces the architecture where it has
to. It leaves some bugs to be fixed later. It adds extensive tests to
verify that unsafe statements work as expected and that appropriate
errors are produced by problems with the selection of binlog format.
It was not practical to split this into smaller pieces of work.
Problem 1:
To determine the logging mode, the code has to take several parameters
into account (namely: (1) the value of binlog_format; (2) the
capabilities of the engines; (3) the type of the current statement:
normal, unsafe, or row injection). These parameters may conflict in
several ways, namely:
- binlog_format=STATEMENT for a row injection
- binlog_format=STATEMENT for an unsafe statement
- binlog_format=STATEMENT for an engine only supporting row logging
- binlog_format=ROW for an engine only supporting statement logging
- statement is unsafe and engine does not support row logging
- row injection in a table that does not support statement logging
- statement modifies one table that does not support row logging and
one that does not support statement logging
Several of these conflicts were not detected, or were detected with
an inappropriate error message. The problem of BUG#39934 was that no
appropriate error message was written for the case when an engine
only supporting row logging executed a row injection with
binlog_format=ROW. However, all above cases must be handled.
Fix 1:
Introduce new error codes (sql/share/errmsg.txt). Ensure that all
conditions are detected and handled in decide_logging_format()
Problem 2:
The binlog format shall be determined once per statement, in
decide_logging_format(). It shall not be changed before or after that.
Before decide_logging_format() is called, all information necessary to
determine the logging format must be available. This principle ensures
that all unsafe statements are handled in a consistent way.
However, this principle is not followed:
thd->set_current_stmt_binlog_row_based_if_mixed() is called in several
places, including from code executing UPDATE..LIMIT,
INSERT..SELECT..LIMIT, DELETE..LIMIT, INSERT DELAYED, and
SET @@binlog_format. After Problem 1 was fixed, that caused
inconsistencies where these unsafe statements would not print the
appropriate warnings or errors for some of the conflicts.
Fix 2:
Remove calls to THD::set_current_stmt_binlog_row_based_if_mixed() from
code executed after decide_logging_format(). Compensate by calling the
set_current_stmt_unsafe() at parse time. This way, all unsafe statements
are detected by decide_logging_format().
Problem 3:
INSERT DELAYED is not unsafe: it is logged in statement format even if
binlog_format=MIXED, and no warning is printed even if
binlog_format=STATEMENT. This is BUG#45825.
Fix 3:
Made INSERT DELAYED set itself to unsafe at parse time. This allows
decide_logging_format() to detect that a warning should be printed or
the binlog_format changed.
Problem 4:
LIMIT clause were not marked as unsafe when executed inside stored
functions/triggers/views/prepared statements. This is
BUG#45785.
Fix 4:
Make statements containing the LIMIT clause marked as unsafe at
parse time, instead of at execution time. This allows propagating
unsafe-ness to the view.
17 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago BUG#39934: Slave stops for engine that only support row-based logging
General overview:
The logic for switching to row format when binlog_format=MIXED had
numerous flaws. The underlying problem was the lack of a consistent
architecture.
General purpose of this changeset:
This changeset introduces an architecture for switching to row format
when binlog_format=MIXED. It enforces the architecture where it has
to. It leaves some bugs to be fixed later. It adds extensive tests to
verify that unsafe statements work as expected and that appropriate
errors are produced by problems with the selection of binlog format.
It was not practical to split this into smaller pieces of work.
Problem 1:
To determine the logging mode, the code has to take several parameters
into account (namely: (1) the value of binlog_format; (2) the
capabilities of the engines; (3) the type of the current statement:
normal, unsafe, or row injection). These parameters may conflict in
several ways, namely:
- binlog_format=STATEMENT for a row injection
- binlog_format=STATEMENT for an unsafe statement
- binlog_format=STATEMENT for an engine only supporting row logging
- binlog_format=ROW for an engine only supporting statement logging
- statement is unsafe and engine does not support row logging
- row injection in a table that does not support statement logging
- statement modifies one table that does not support row logging and
one that does not support statement logging
Several of these conflicts were not detected, or were detected with
an inappropriate error message. The problem of BUG#39934 was that no
appropriate error message was written for the case when an engine
only supporting row logging executed a row injection with
binlog_format=ROW. However, all above cases must be handled.
Fix 1:
Introduce new error codes (sql/share/errmsg.txt). Ensure that all
conditions are detected and handled in decide_logging_format()
Problem 2:
The binlog format shall be determined once per statement, in
decide_logging_format(). It shall not be changed before or after that.
Before decide_logging_format() is called, all information necessary to
determine the logging format must be available. This principle ensures
that all unsafe statements are handled in a consistent way.
However, this principle is not followed:
thd->set_current_stmt_binlog_row_based_if_mixed() is called in several
places, including from code executing UPDATE..LIMIT,
INSERT..SELECT..LIMIT, DELETE..LIMIT, INSERT DELAYED, and
SET @@binlog_format. After Problem 1 was fixed, that caused
inconsistencies where these unsafe statements would not print the
appropriate warnings or errors for some of the conflicts.
Fix 2:
Remove calls to THD::set_current_stmt_binlog_row_based_if_mixed() from
code executed after decide_logging_format(). Compensate by calling the
set_current_stmt_unsafe() at parse time. This way, all unsafe statements
are detected by decide_logging_format().
Problem 3:
INSERT DELAYED is not unsafe: it is logged in statement format even if
binlog_format=MIXED, and no warning is printed even if
binlog_format=STATEMENT. This is BUG#45825.
Fix 3:
Made INSERT DELAYED set itself to unsafe at parse time. This allows
decide_logging_format() to detect that a warning should be printed or
the binlog_format changed.
Problem 4:
LIMIT clause were not marked as unsafe when executed inside stored
functions/triggers/views/prepared statements. This is
BUG#45785.
Fix 4:
Make statements containing the LIMIT clause marked as unsafe at
parse time, instead of at execution time. This allows propagating
unsafe-ness to the view.
17 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
20 years ago BUG#22864 (Rollback following CREATE... SELECT discards 'CREATE TABLE'
from log):
When row-based logging is used, the CREATE-SELECT is written as two
parts: as a CREATE TABLE statement and as the rows for the table. For
both transactional and non-transactional tables, the CREATE TABLE
statement was written to the transaction cache, as were the rows, and
on statement end, the entire transaction cache was written to the binary
log if the table was non-transactional. For transactional tables, the
events were kept in the transaction cache until end of transaction (or
statement that were not part of a transaction).
For the case when AUTOCOMMIT=0 and we are creating a transactional table
using a create select, we would then keep the CREATE TABLE statement and
the rows for the CREATE-SELECT, while executing the following statements.
On a rollback, the transaction cache would then be cleared, which would
also remove the CREATE TABLE statement. Hence no table would be created
on the slave, while there is an empty table on the master.
This relates to BUG#22865 where the table being created exists on the
master, but not on the slave during insertion of rows into the newly
created table. This occurs since the CREATE TABLE statement were still
in the transaction cache until the statement finished executing, and
possibly longer if the table was transactional.
This patch changes the behaviour of the CREATE-SELECT statement by
adding an implicit commit at the end of the statement when creating
non-temporary tables. Hence, non-temporary tables will be written to the
binary log on completion, and in the even of AUTOCOMMIT=0, a new
transaction will be started. Temporary tables do not commit an ongoing
transaction: neither as a pre- not a post-commit.
The events for both transactional and non-transactional tables are
saved in the transaction cache, and written to the binary log at end
of the statement.
19 years ago BUG#39934: Slave stops for engine that only support row-based logging
General overview:
The logic for switching to row format when binlog_format=MIXED had
numerous flaws. The underlying problem was the lack of a consistent
architecture.
General purpose of this changeset:
This changeset introduces an architecture for switching to row format
when binlog_format=MIXED. It enforces the architecture where it has
to. It leaves some bugs to be fixed later. It adds extensive tests to
verify that unsafe statements work as expected and that appropriate
errors are produced by problems with the selection of binlog format.
It was not practical to split this into smaller pieces of work.
Problem 1:
To determine the logging mode, the code has to take several parameters
into account (namely: (1) the value of binlog_format; (2) the
capabilities of the engines; (3) the type of the current statement:
normal, unsafe, or row injection). These parameters may conflict in
several ways, namely:
- binlog_format=STATEMENT for a row injection
- binlog_format=STATEMENT for an unsafe statement
- binlog_format=STATEMENT for an engine only supporting row logging
- binlog_format=ROW for an engine only supporting statement logging
- statement is unsafe and engine does not support row logging
- row injection in a table that does not support statement logging
- statement modifies one table that does not support row logging and
one that does not support statement logging
Several of these conflicts were not detected, or were detected with
an inappropriate error message. The problem of BUG#39934 was that no
appropriate error message was written for the case when an engine
only supporting row logging executed a row injection with
binlog_format=ROW. However, all above cases must be handled.
Fix 1:
Introduce new error codes (sql/share/errmsg.txt). Ensure that all
conditions are detected and handled in decide_logging_format()
Problem 2:
The binlog format shall be determined once per statement, in
decide_logging_format(). It shall not be changed before or after that.
Before decide_logging_format() is called, all information necessary to
determine the logging format must be available. This principle ensures
that all unsafe statements are handled in a consistent way.
However, this principle is not followed:
thd->set_current_stmt_binlog_row_based_if_mixed() is called in several
places, including from code executing UPDATE..LIMIT,
INSERT..SELECT..LIMIT, DELETE..LIMIT, INSERT DELAYED, and
SET @@binlog_format. After Problem 1 was fixed, that caused
inconsistencies where these unsafe statements would not print the
appropriate warnings or errors for some of the conflicts.
Fix 2:
Remove calls to THD::set_current_stmt_binlog_row_based_if_mixed() from
code executed after decide_logging_format(). Compensate by calling the
set_current_stmt_unsafe() at parse time. This way, all unsafe statements
are detected by decide_logging_format().
Problem 3:
INSERT DELAYED is not unsafe: it is logged in statement format even if
binlog_format=MIXED, and no warning is printed even if
binlog_format=STATEMENT. This is BUG#45825.
Fix 3:
Made INSERT DELAYED set itself to unsafe at parse time. This allows
decide_logging_format() to detect that a warning should be printed or
the binlog_format changed.
Problem 4:
LIMIT clause were not marked as unsafe when executed inside stored
functions/triggers/views/prepared statements. This is
BUG#45785.
Fix 4:
Make statements containing the LIMIT clause marked as unsafe at
parse time, instead of at execution time. This allows propagating
unsafe-ness to the view.
17 years ago BUG#22864 (Rollback following CREATE... SELECT discards 'CREATE TABLE'
from log):
When row-based logging is used, the CREATE-SELECT is written as two
parts: as a CREATE TABLE statement and as the rows for the table. For
both transactional and non-transactional tables, the CREATE TABLE
statement was written to the transaction cache, as were the rows, and
on statement end, the entire transaction cache was written to the binary
log if the table was non-transactional. For transactional tables, the
events were kept in the transaction cache until end of transaction (or
statement that were not part of a transaction).
For the case when AUTOCOMMIT=0 and we are creating a transactional table
using a create select, we would then keep the CREATE TABLE statement and
the rows for the CREATE-SELECT, while executing the following statements.
On a rollback, the transaction cache would then be cleared, which would
also remove the CREATE TABLE statement. Hence no table would be created
on the slave, while there is an empty table on the master.
This relates to BUG#22865 where the table being created exists on the
master, but not on the slave during insertion of rows into the newly
created table. This occurs since the CREATE TABLE statement were still
in the transaction cache until the statement finished executing, and
possibly longer if the table was transactional.
This patch changes the behaviour of the CREATE-SELECT statement by
adding an implicit commit at the end of the statement when creating
non-temporary tables. Hence, non-temporary tables will be written to the
binary log on completion, and in the even of AUTOCOMMIT=0, a new
transaction will be started. Temporary tables do not commit an ongoing
transaction: neither as a pre- not a post-commit.
The events for both transactional and non-transactional tables are
saved in the transaction cache, and written to the binary log at end
of the statement.
19 years ago BUG#22864 (Rollback following CREATE... SELECT discards 'CREATE TABLE'
from log):
When row-based logging is used, the CREATE-SELECT is written as two
parts: as a CREATE TABLE statement and as the rows for the table. For
both transactional and non-transactional tables, the CREATE TABLE
statement was written to the transaction cache, as were the rows, and
on statement end, the entire transaction cache was written to the binary
log if the table was non-transactional. For transactional tables, the
events were kept in the transaction cache until end of transaction (or
statement that were not part of a transaction).
For the case when AUTOCOMMIT=0 and we are creating a transactional table
using a create select, we would then keep the CREATE TABLE statement and
the rows for the CREATE-SELECT, while executing the following statements.
On a rollback, the transaction cache would then be cleared, which would
also remove the CREATE TABLE statement. Hence no table would be created
on the slave, while there is an empty table on the master.
This relates to BUG#22865 where the table being created exists on the
master, but not on the slave during insertion of rows into the newly
created table. This occurs since the CREATE TABLE statement were still
in the transaction cache until the statement finished executing, and
possibly longer if the table was transactional.
This patch changes the behaviour of the CREATE-SELECT statement by
adding an implicit commit at the end of the statement when creating
non-temporary tables. Hence, non-temporary tables will be written to the
binary log on completion, and in the even of AUTOCOMMIT=0, a new
transaction will be started. Temporary tables do not commit an ongoing
transaction: neither as a pre- not a post-commit.
The events for both transactional and non-transactional tables are
saved in the transaction cache, and written to the binary log at end
of the statement.
19 years ago BUG#22864 (Rollback following CREATE... SELECT discards 'CREATE TABLE'
from log):
When row-based logging is used, the CREATE-SELECT is written as two
parts: as a CREATE TABLE statement and as the rows for the table. For
both transactional and non-transactional tables, the CREATE TABLE
statement was written to the transaction cache, as were the rows, and
on statement end, the entire transaction cache was written to the binary
log if the table was non-transactional. For transactional tables, the
events were kept in the transaction cache until end of transaction (or
statement that were not part of a transaction).
For the case when AUTOCOMMIT=0 and we are creating a transactional table
using a create select, we would then keep the CREATE TABLE statement and
the rows for the CREATE-SELECT, while executing the following statements.
On a rollback, the transaction cache would then be cleared, which would
also remove the CREATE TABLE statement. Hence no table would be created
on the slave, while there is an empty table on the master.
This relates to BUG#22865 where the table being created exists on the
master, but not on the slave during insertion of rows into the newly
created table. This occurs since the CREATE TABLE statement were still
in the transaction cache until the statement finished executing, and
possibly longer if the table was transactional.
This patch changes the behaviour of the CREATE-SELECT statement by
adding an implicit commit at the end of the statement when creating
non-temporary tables. Hence, non-temporary tables will be written to the
binary log on completion, and in the even of AUTOCOMMIT=0, a new
transaction will be started. Temporary tables do not commit an ongoing
transaction: neither as a pre- not a post-commit.
The events for both transactional and non-transactional tables are
saved in the transaction cache, and written to the binary log at end
of the statement.
19 years ago BUG#22864 (Rollback following CREATE... SELECT discards 'CREATE TABLE'
from log):
When row-based logging is used, the CREATE-SELECT is written as two
parts: as a CREATE TABLE statement and as the rows for the table. For
both transactional and non-transactional tables, the CREATE TABLE
statement was written to the transaction cache, as were the rows, and
on statement end, the entire transaction cache was written to the binary
log if the table was non-transactional. For transactional tables, the
events were kept in the transaction cache until end of transaction (or
statement that were not part of a transaction).
For the case when AUTOCOMMIT=0 and we are creating a transactional table
using a create select, we would then keep the CREATE TABLE statement and
the rows for the CREATE-SELECT, while executing the following statements.
On a rollback, the transaction cache would then be cleared, which would
also remove the CREATE TABLE statement. Hence no table would be created
on the slave, while there is an empty table on the master.
This relates to BUG#22865 where the table being created exists on the
master, but not on the slave during insertion of rows into the newly
created table. This occurs since the CREATE TABLE statement were still
in the transaction cache until the statement finished executing, and
possibly longer if the table was transactional.
This patch changes the behaviour of the CREATE-SELECT statement by
adding an implicit commit at the end of the statement when creating
non-temporary tables. Hence, non-temporary tables will be written to the
binary log on completion, and in the even of AUTOCOMMIT=0, a new
transaction will be started. Temporary tables do not commit an ongoing
transaction: neither as a pre- not a post-commit.
The events for both transactional and non-transactional tables are
saved in the transaction cache, and written to the binary log at end
of the statement.
19 years ago BUG#22864 (Rollback following CREATE... SELECT discards 'CREATE TABLE'
from log):
When row-based logging is used, the CREATE-SELECT is written as two
parts: as a CREATE TABLE statement and as the rows for the table. For
both transactional and non-transactional tables, the CREATE TABLE
statement was written to the transaction cache, as were the rows, and
on statement end, the entire transaction cache was written to the binary
log if the table was non-transactional. For transactional tables, the
events were kept in the transaction cache until end of transaction (or
statement that were not part of a transaction).
For the case when AUTOCOMMIT=0 and we are creating a transactional table
using a create select, we would then keep the CREATE TABLE statement and
the rows for the CREATE-SELECT, while executing the following statements.
On a rollback, the transaction cache would then be cleared, which would
also remove the CREATE TABLE statement. Hence no table would be created
on the slave, while there is an empty table on the master.
This relates to BUG#22865 where the table being created exists on the
master, but not on the slave during insertion of rows into the newly
created table. This occurs since the CREATE TABLE statement were still
in the transaction cache until the statement finished executing, and
possibly longer if the table was transactional.
This patch changes the behaviour of the CREATE-SELECT statement by
adding an implicit commit at the end of the statement when creating
non-temporary tables. Hence, non-temporary tables will be written to the
binary log on completion, and in the even of AUTOCOMMIT=0, a new
transaction will be started. Temporary tables do not commit an ongoing
transaction: neither as a pre- not a post-commit.
The events for both transactional and non-transactional tables are
saved in the transaction cache, and written to the binary log at end
of the statement.
19 years ago BUG#22864 (Rollback following CREATE... SELECT discards 'CREATE TABLE'
from log):
When row-based logging is used, the CREATE-SELECT is written as two
parts: as a CREATE TABLE statement and as the rows for the table. For
both transactional and non-transactional tables, the CREATE TABLE
statement was written to the transaction cache, as were the rows, and
on statement end, the entire transaction cache was written to the binary
log if the table was non-transactional. For transactional tables, the
events were kept in the transaction cache until end of transaction (or
statement that were not part of a transaction).
For the case when AUTOCOMMIT=0 and we are creating a transactional table
using a create select, we would then keep the CREATE TABLE statement and
the rows for the CREATE-SELECT, while executing the following statements.
On a rollback, the transaction cache would then be cleared, which would
also remove the CREATE TABLE statement. Hence no table would be created
on the slave, while there is an empty table on the master.
This relates to BUG#22865 where the table being created exists on the
master, but not on the slave during insertion of rows into the newly
created table. This occurs since the CREATE TABLE statement were still
in the transaction cache until the statement finished executing, and
possibly longer if the table was transactional.
This patch changes the behaviour of the CREATE-SELECT statement by
adding an implicit commit at the end of the statement when creating
non-temporary tables. Hence, non-temporary tables will be written to the
binary log on completion, and in the even of AUTOCOMMIT=0, a new
transaction will be started. Temporary tables do not commit an ongoing
transaction: neither as a pre- not a post-commit.
The events for both transactional and non-transactional tables are
saved in the transaction cache, and written to the binary log at end
of the statement.
19 years ago BUG#22864 (Rollback following CREATE... SELECT discards 'CREATE TABLE'
from log):
When row-based logging is used, the CREATE-SELECT is written as two
parts: as a CREATE TABLE statement and as the rows for the table. For
both transactional and non-transactional tables, the CREATE TABLE
statement was written to the transaction cache, as were the rows, and
on statement end, the entire transaction cache was written to the binary
log if the table was non-transactional. For transactional tables, the
events were kept in the transaction cache until end of transaction (or
statement that were not part of a transaction).
For the case when AUTOCOMMIT=0 and we are creating a transactional table
using a create select, we would then keep the CREATE TABLE statement and
the rows for the CREATE-SELECT, while executing the following statements.
On a rollback, the transaction cache would then be cleared, which would
also remove the CREATE TABLE statement. Hence no table would be created
on the slave, while there is an empty table on the master.
This relates to BUG#22865 where the table being created exists on the
master, but not on the slave during insertion of rows into the newly
created table. This occurs since the CREATE TABLE statement were still
in the transaction cache until the statement finished executing, and
possibly longer if the table was transactional.
This patch changes the behaviour of the CREATE-SELECT statement by
adding an implicit commit at the end of the statement when creating
non-temporary tables. Hence, non-temporary tables will be written to the
binary log on completion, and in the even of AUTOCOMMIT=0, a new
transaction will be started. Temporary tables do not commit an ongoing
transaction: neither as a pre- not a post-commit.
The events for both transactional and non-transactional tables are
saved in the transaction cache, and written to the binary log at end
of the statement.
19 years ago BUG#22864 (Rollback following CREATE... SELECT discards 'CREATE TABLE'
from log):
When row-based logging is used, the CREATE-SELECT is written as two
parts: as a CREATE TABLE statement and as the rows for the table. For
both transactional and non-transactional tables, the CREATE TABLE
statement was written to the transaction cache, as were the rows, and
on statement end, the entire transaction cache was written to the binary
log if the table was non-transactional. For transactional tables, the
events were kept in the transaction cache until end of transaction (or
statement that were not part of a transaction).
For the case when AUTOCOMMIT=0 and we are creating a transactional table
using a create select, we would then keep the CREATE TABLE statement and
the rows for the CREATE-SELECT, while executing the following statements.
On a rollback, the transaction cache would then be cleared, which would
also remove the CREATE TABLE statement. Hence no table would be created
on the slave, while there is an empty table on the master.
This relates to BUG#22865 where the table being created exists on the
master, but not on the slave during insertion of rows into the newly
created table. This occurs since the CREATE TABLE statement were still
in the transaction cache until the statement finished executing, and
possibly longer if the table was transactional.
This patch changes the behaviour of the CREATE-SELECT statement by
adding an implicit commit at the end of the statement when creating
non-temporary tables. Hence, non-temporary tables will be written to the
binary log on completion, and in the even of AUTOCOMMIT=0, a new
transaction will be started. Temporary tables do not commit an ongoing
transaction: neither as a pre- not a post-commit.
The events for both transactional and non-transactional tables are
saved in the transaction cache, and written to the binary log at end
of the statement.
19 years ago BUG#22864 (Rollback following CREATE... SELECT discards 'CREATE TABLE'
from log):
When row-based logging is used, the CREATE-SELECT is written as two
parts: as a CREATE TABLE statement and as the rows for the table. For
both transactional and non-transactional tables, the CREATE TABLE
statement was written to the transaction cache, as were the rows, and
on statement end, the entire transaction cache was written to the binary
log if the table was non-transactional. For transactional tables, the
events were kept in the transaction cache until end of transaction (or
statement that were not part of a transaction).
For the case when AUTOCOMMIT=0 and we are creating a transactional table
using a create select, we would then keep the CREATE TABLE statement and
the rows for the CREATE-SELECT, while executing the following statements.
On a rollback, the transaction cache would then be cleared, which would
also remove the CREATE TABLE statement. Hence no table would be created
on the slave, while there is an empty table on the master.
This relates to BUG#22865 where the table being created exists on the
master, but not on the slave during insertion of rows into the newly
created table. This occurs since the CREATE TABLE statement were still
in the transaction cache until the statement finished executing, and
possibly longer if the table was transactional.
This patch changes the behaviour of the CREATE-SELECT statement by
adding an implicit commit at the end of the statement when creating
non-temporary tables. Hence, non-temporary tables will be written to the
binary log on completion, and in the even of AUTOCOMMIT=0, a new
transaction will be started. Temporary tables do not commit an ongoing
transaction: neither as a pre- not a post-commit.
The events for both transactional and non-transactional tables are
saved in the transaction cache, and written to the binary log at end
of the statement.
19 years ago BUG#22864 (Rollback following CREATE... SELECT discards 'CREATE TABLE'
from log):
When row-based logging is used, the CREATE-SELECT is written as two
parts: as a CREATE TABLE statement and as the rows for the table. For
both transactional and non-transactional tables, the CREATE TABLE
statement was written to the transaction cache, as were the rows, and
on statement end, the entire transaction cache was written to the binary
log if the table was non-transactional. For transactional tables, the
events were kept in the transaction cache until end of transaction (or
statement that were not part of a transaction).
For the case when AUTOCOMMIT=0 and we are creating a transactional table
using a create select, we would then keep the CREATE TABLE statement and
the rows for the CREATE-SELECT, while executing the following statements.
On a rollback, the transaction cache would then be cleared, which would
also remove the CREATE TABLE statement. Hence no table would be created
on the slave, while there is an empty table on the master.
This relates to BUG#22865 where the table being created exists on the
master, but not on the slave during insertion of rows into the newly
created table. This occurs since the CREATE TABLE statement were still
in the transaction cache until the statement finished executing, and
possibly longer if the table was transactional.
This patch changes the behaviour of the CREATE-SELECT statement by
adding an implicit commit at the end of the statement when creating
non-temporary tables. Hence, non-temporary tables will be written to the
binary log on completion, and in the even of AUTOCOMMIT=0, a new
transaction will be started. Temporary tables do not commit an ongoing
transaction: neither as a pre- not a post-commit.
The events for both transactional and non-transactional tables are
saved in the transaction cache, and written to the binary log at end
of the statement.
19 years ago BUG#22864 (Rollback following CREATE... SELECT discards 'CREATE TABLE'
from log):
When row-based logging is used, the CREATE-SELECT is written as two
parts: as a CREATE TABLE statement and as the rows for the table. For
both transactional and non-transactional tables, the CREATE TABLE
statement was written to the transaction cache, as were the rows, and
on statement end, the entire transaction cache was written to the binary
log if the table was non-transactional. For transactional tables, the
events were kept in the transaction cache until end of transaction (or
statement that were not part of a transaction).
For the case when AUTOCOMMIT=0 and we are creating a transactional table
using a create select, we would then keep the CREATE TABLE statement and
the rows for the CREATE-SELECT, while executing the following statements.
On a rollback, the transaction cache would then be cleared, which would
also remove the CREATE TABLE statement. Hence no table would be created
on the slave, while there is an empty table on the master.
This relates to BUG#22865 where the table being created exists on the
master, but not on the slave during insertion of rows into the newly
created table. This occurs since the CREATE TABLE statement were still
in the transaction cache until the statement finished executing, and
possibly longer if the table was transactional.
This patch changes the behaviour of the CREATE-SELECT statement by
adding an implicit commit at the end of the statement when creating
non-temporary tables. Hence, non-temporary tables will be written to the
binary log on completion, and in the even of AUTOCOMMIT=0, a new
transaction will be started. Temporary tables do not commit an ongoing
transaction: neither as a pre- not a post-commit.
The events for both transactional and non-transactional tables are
saved in the transaction cache, and written to the binary log at end
of the statement.
19 years ago |
|
/* Copyright 2000-2008 MySQL AB, 2008-2009 Sun Microsystems, Inc.
This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; version 2 of the License.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */
/* Insert of records */
/*
INSERT DELAYED
Insert delayed is distinguished from a normal insert by lock_type == TL_WRITE_DELAYED instead of TL_WRITE. It first tries to open a "delayed" table (delayed_get_table()), but falls back to open_and_lock_tables() on error and proceeds as normal insert then.
Opening a "delayed" table means to find a delayed insert thread that has the table open already. If this fails, a new thread is created and waited for to open and lock the table.
If accessing the thread succeeded, in Delayed_insert::get_local_table() the table of the thread is copied for local use. A copy is required because the normal insert logic works on a target table, but the other threads table object must not be used. The insert logic uses the record buffer to create a record. And the delayed insert thread uses the record buffer to pass the record to the table handler. So there must be different objects. Also the copied table is not included in the lock, so that the statement can proceed even if the real table cannot be accessed at this moment.
Copying a table object is not a trivial operation. Besides the TABLE object there are the field pointer array, the field objects and the record buffer. After copying the field objects, their pointers into the record must be "moved" to point to the new record buffer.
After this setup the normal insert logic is used. Only that for delayed inserts write_delayed() is called instead of write_record(). It inserts the rows into a queue and signals the delayed insert thread instead of writing directly to the table.
The delayed insert thread awakes from the signal. It locks the table, inserts the rows from the queue, unlocks the table, and waits for the next signal. It does normally live until a FLUSH TABLES or SHUTDOWN.
*/
#include "my_global.h" /* NO_EMBEDDED_ACCESS_CHECKS */
#include "sql_priv.h"
#include "unireg.h" // REQUIRED: for other includes
#include "sql_insert.h"
#include "sql_update.h" // compare_record
#include "sql_base.h" // close_thread_tables
#include "sql_cache.h" // query_cache_*
#include "key.h" // key_copy
#include "lock.h" // mysql_unlock_tables
#include "sp_head.h"
#include "sql_view.h" // check_key_in_view, insert_view_fields
#include "sql_table.h" // mysql_create_table_no_lock
#include "sql_acl.h" // *_ACL, check_grant_all_columns
#include "sql_trigger.h"
#include "sql_select.h"
#include "sql_show.h"
#include "slave.h"
#include "sql_parse.h" // end_active_trans
#include "rpl_mi.h"
#include "transaction.h"
#include "sql_audit.h"
#ifndef EMBEDDED_LIBRARY
static bool delayed_get_table(THD *thd, TABLE_LIST *table_list);static int write_delayed(THD *thd, TABLE *table, enum_duplicates duplic, LEX_STRING query, bool ignore, bool log_on);static void end_delayed_insert(THD *thd);pthread_handler_t handle_delayed_insert(void *arg);static void unlink_blobs(register TABLE *table);#endif
static bool check_view_insertability(THD *thd, TABLE_LIST *view);
/* Define to force use of my_malloc() if the allocated memory block is big */
#ifndef HAVE_ALLOCA
#define my_safe_alloca(size, min_length) my_alloca(size)
#define my_safe_afree(ptr, size, min_length) my_afree(ptr)
#else
#define my_safe_alloca(size, min_length) ((size <= min_length) ? my_alloca(size) : my_malloc(size,MYF(0)))
#define my_safe_afree(ptr, size, min_length) if (size > min_length) my_free(ptr)
#endif
/*
Check that insert/update fields are from the same single table of a view.
SYNOPSIS check_view_single_update() fields The insert/update fields to be checked. view The view for insert. map [in/out] The insert table map.
DESCRIPTION This function is called in 2 cases: 1. to check insert fields. In this case *map will be set to 0. Insert fields are checked to be all from the same single underlying table of the given view. Otherwise the error is thrown. Found table map is returned in the map parameter. 2. to check update fields of the ON DUPLICATE KEY UPDATE clause. In this case *map contains table_map found on the previous call of the function to check insert fields. Update fields are checked to be from the same table as the insert fields.
RETURN 0 OK 1 Error*/
bool check_view_single_update(List<Item> &fields, List<Item> *values, TABLE_LIST *view, table_map *map){ /* it is join view => we need to find the table for update */ List_iterator_fast<Item> it(fields); Item *item; TABLE_LIST *tbl= 0; // reset for call to check_single_table()
table_map tables= 0;
while ((item= it++)) tables|= item->used_tables();
if (values) { it.init(*values); while ((item= it++)) tables|= item->used_tables(); }
/* Convert to real table bits */ tables&= ~PSEUDO_TABLE_BITS;
/* Check found map against provided map */ if (*map) { if (tables != *map) goto error; return FALSE; }
if (view->check_single_table(&tbl, tables, view) || tbl == 0) goto error;
view->table= tbl->table; *map= tables;
return FALSE;
error: my_error(ER_VIEW_MULTIUPDATE, MYF(0), view->view_db.str, view->view_name.str); return TRUE;}
/*
Check if insert fields are correct.
SYNOPSIS check_insert_fields() thd The current thread. table The table for insert. fields The insert fields. values The insert values. check_unique If duplicate values should be rejected.
NOTE Clears TIMESTAMP_AUTO_SET_ON_INSERT from table->timestamp_field_type or leaves it as is, depending on if timestamp should be updated or not.
RETURN 0 OK -1 Error*/
static int check_insert_fields(THD *thd, TABLE_LIST *table_list, List<Item> &fields, List<Item> &values, bool check_unique, bool fields_and_values_from_different_maps, table_map *map){ TABLE *table= table_list->table;
if (!table_list->updatable) { my_error(ER_NON_INSERTABLE_TABLE, MYF(0), table_list->alias, "INSERT"); return -1; }
if (fields.elements == 0 && values.elements != 0) { if (!table) { my_error(ER_VIEW_NO_INSERT_FIELD_LIST, MYF(0), table_list->view_db.str, table_list->view_name.str); return -1; } if (values.elements != table->s->fields) { my_error(ER_WRONG_VALUE_COUNT_ON_ROW, MYF(0), 1L); return -1; }#ifndef NO_EMBEDDED_ACCESS_CHECKS
Field_iterator_table_ref field_it; field_it.set(table_list); if (check_grant_all_columns(thd, INSERT_ACL, &field_it)) return -1;#endif
clear_timestamp_auto_bits(table->timestamp_field_type, TIMESTAMP_AUTO_SET_ON_INSERT); /*
No fields are provided so all fields must be provided in the values. Thus we set all bits in the write set. */ bitmap_set_all(table->write_set); } else { // Part field list
SELECT_LEX *select_lex= &thd->lex->select_lex; Name_resolution_context *context= &select_lex->context; Name_resolution_context_state ctx_state; int res;
if (fields.elements != values.elements) { my_error(ER_WRONG_VALUE_COUNT_ON_ROW, MYF(0), 1L); return -1; }
thd->dup_field= 0; select_lex->no_wrap_view_item= TRUE;
/* Save the state of the current name resolution context. */ ctx_state.save_state(context, table_list);
/*
Perform name resolution only in the first table - 'table_list', which is the table that is inserted into. */ table_list->next_local= 0; context->resolve_in_table_list_only(table_list); res= setup_fields(thd, 0, fields, MARK_COLUMNS_WRITE, 0, 0);
/* Restore the current context. */ ctx_state.restore_state(context, table_list); thd->lex->select_lex.no_wrap_view_item= FALSE;
if (res) return -1;
if (table_list->effective_algorithm == VIEW_ALGORITHM_MERGE) { if (check_view_single_update(fields, fields_and_values_from_different_maps ? (List<Item>*) 0 : &values, table_list, map)) return -1; table= table_list->table; }
if (check_unique && thd->dup_field) { my_error(ER_FIELD_SPECIFIED_TWICE, MYF(0), thd->dup_field->field_name); return -1; } if (table->timestamp_field) // Don't automaticly set timestamp if used
{ if (bitmap_is_set(table->write_set, table->timestamp_field->field_index)) clear_timestamp_auto_bits(table->timestamp_field_type, TIMESTAMP_AUTO_SET_ON_INSERT); else { bitmap_set_bit(table->write_set, table->timestamp_field->field_index); } } } // For the values we need select_priv
#ifndef NO_EMBEDDED_ACCESS_CHECKS
table->grant.want_privilege= (SELECT_ACL & ~table->grant.privilege);#endif
if (check_key_in_view(thd, table_list) || (table_list->view && check_view_insertability(thd, table_list))) { my_error(ER_NON_INSERTABLE_TABLE, MYF(0), table_list->alias, "INSERT"); return -1; }
return 0;}
/*
Check update fields for the timestamp field.
SYNOPSIS check_update_fields() thd The current thread. insert_table_list The insert table list. table The table for update. update_fields The update fields.
NOTE If the update fields include the timestamp field, remove TIMESTAMP_AUTO_SET_ON_UPDATE from table->timestamp_field_type.
RETURN 0 OK -1 Error*/
static int check_update_fields(THD *thd, TABLE_LIST *insert_table_list, List<Item> &update_fields, List<Item> &update_values, table_map *map){ TABLE *table= insert_table_list->table; my_bool timestamp_mark= 0;
if (table->timestamp_field) { /*
Unmark the timestamp field so that we can check if this is modified by update_fields */ timestamp_mark= bitmap_test_and_clear(table->write_set, table->timestamp_field->field_index); }
/* Check the fields we are going to modify */ if (setup_fields(thd, 0, update_fields, MARK_COLUMNS_WRITE, 0, 0)) return -1;
if (insert_table_list->effective_algorithm == VIEW_ALGORITHM_MERGE && check_view_single_update(update_fields, &update_values, insert_table_list, map)) return -1;
if (table->timestamp_field) { /* Don't set timestamp column if this is modified. */ if (bitmap_is_set(table->write_set, table->timestamp_field->field_index)) clear_timestamp_auto_bits(table->timestamp_field_type, TIMESTAMP_AUTO_SET_ON_UPDATE); if (timestamp_mark) bitmap_set_bit(table->write_set, table->timestamp_field->field_index); } return 0;}
/*
Prepare triggers for INSERT-like statement.
SYNOPSIS prepare_triggers_for_insert_stmt() table Table to which insert will happen
NOTE Prepare triggers for INSERT-like statement by marking fields used by triggers and inform handlers that batching of UPDATE/DELETE cannot be done if there are BEFORE UPDATE/DELETE triggers.*/
void prepare_triggers_for_insert_stmt(TABLE *table){ if (table->triggers) { if (table->triggers->has_triggers(TRG_EVENT_DELETE, TRG_ACTION_AFTER)) { /*
The table has AFTER DELETE triggers that might access to subject table and therefore might need delete to be done immediately. So we turn-off the batching. */ (void) table->file->extra(HA_EXTRA_DELETE_CANNOT_BATCH); } if (table->triggers->has_triggers(TRG_EVENT_UPDATE, TRG_ACTION_AFTER)) { /*
The table has AFTER UPDATE triggers that might access to subject table and therefore might need update to be done immediately. So we turn-off the batching. */ (void) table->file->extra(HA_EXTRA_UPDATE_CANNOT_BATCH); } } table->mark_columns_needed_for_insert();}
/**
Upgrade table-level lock of INSERT statement to TL_WRITE if a more concurrent lock is infeasible for some reason. This is necessary for engines without internal locking support (MyISAM). An engine with internal locking implementation might later downgrade the lock in handler::store_lock() method.*/
staticvoid upgrade_lock_type(THD *thd, thr_lock_type *lock_type, enum_duplicates duplic, bool is_multi_insert){ if (duplic == DUP_UPDATE || (duplic == DUP_REPLACE && *lock_type == TL_WRITE_CONCURRENT_INSERT)) { *lock_type= TL_WRITE_DEFAULT; return; }
if (*lock_type == TL_WRITE_DELAYED) { /*
We do not use delayed threads if: - we're running in the safe mode or skip-new mode -- the feature is disabled in these modes - we're executing this statement on a replication slave -- we need to ensure serial execution of queries on the slave - it is INSERT .. ON DUPLICATE KEY UPDATE - in this case the insert cannot be concurrent - this statement is directly or indirectly invoked from a stored function or trigger (under pre-locking) - to avoid deadlocks, since INSERT DELAYED involves a lock upgrade (TL_WRITE_DELAYED -> TL_WRITE) which we should not attempt while keeping other table level locks. - this statement itself may require pre-locking. We should upgrade the lock even though in most cases delayed functionality may work. Unfortunately, we can't easily identify whether the subject table is not used in the statement indirectly via a stored function or trigger: if it is used, that will lead to a deadlock between the client connection and the delayed thread. */ if (specialflag & (SPECIAL_NO_NEW_FUNC | SPECIAL_SAFE_MODE) || thd->variables.max_insert_delayed_threads == 0 || thd->locked_tables_mode > LTM_LOCK_TABLES || thd->lex->uses_stored_routines()) { *lock_type= TL_WRITE; return; } if (thd->slave_thread) { /* Try concurrent insert */ *lock_type= (duplic == DUP_UPDATE || duplic == DUP_REPLACE) ? TL_WRITE : TL_WRITE_CONCURRENT_INSERT; return; }
bool log_on= (thd->variables.option_bits & OPTION_BIN_LOG || ! (thd->security_ctx->master_access & SUPER_ACL)); if (global_system_variables.binlog_format == BINLOG_FORMAT_STMT && log_on && mysql_bin_log.is_open() && is_multi_insert) { /*
Statement-based binary logging does not work in this case, because: a) two concurrent statements may have their rows intermixed in the queue, leading to autoincrement replication problems on slave (because the values generated used for one statement don't depend only on the value generated for the first row of this statement, so are not replicable) b) if first row of the statement has an error the full statement is not binlogged, while next rows of the statement may be inserted. c) if first row succeeds, statement is binlogged immediately with a zero error code (i.e. "no error"), if then second row fails, query will fail on slave too and slave will stop (wrongly believing that the master got no error). So we fallback to non-delayed INSERT. Note that to be fully correct, we should test the "binlog format which the delayed thread is going to use for this row". But in the common case where the global binlog format is not changed and the session binlog format may be changed, that is equal to the global binlog format. We test it without mutex for speed reasons (condition rarely true), and in the common case (global not changed) it is as good as without mutex; if global value is changed, anyway there is uncertainty as the delayed thread may be old and use the before-the-change value. */ *lock_type= TL_WRITE; } }}
/**
Find or create a delayed insert thread for the first table in the table list, then open and lock the remaining tables. If a table can not be used with insert delayed, upgrade the lock and open and lock all tables using the standard mechanism.
@param thd thread context @param table_list list of "descriptors" for tables referenced directly in statement SQL text. The first element in the list corresponds to the destination table for inserts, remaining tables, if any, are usually tables referenced by sub-queries in the right part of the INSERT.
@return Status of the operation. In case of success 'table' member of every table_list element points to an instance of class TABLE.
@sa open_and_lock_tables for more information about MySQL table level locking*/
staticbool open_and_lock_for_insert_delayed(THD *thd, TABLE_LIST *table_list){ DBUG_ENTER("open_and_lock_for_insert_delayed");
#ifndef EMBEDDED_LIBRARY
if (thd->locked_tables_mode && thd->global_read_lock.is_acquired()) { /*
If this connection has the global read lock, the handler thread will not be able to lock the table. It will wait for the global read lock to go away, but this will never happen since the connection thread will be stuck waiting for the handler thread to open and lock the table. If we are not in locked tables mode, INSERT will seek protection against the global read lock (and fail), thus we will only get to this point in locked tables mode. */ my_error(ER_CANT_UPDATE_WITH_READLOCK, MYF(0)); DBUG_RETURN(TRUE); }
/*
In order for the deadlock detector to be able to find any deadlocks caused by the handler thread locking this table, we take the metadata lock inside the connection thread. If this goes ok, the ticket is cloned and added to the list of granted locks held by the handler thread. */ MDL_ticket *mdl_savepoint= thd->mdl_context.mdl_savepoint(); if (thd->mdl_context.acquire_lock(&table_list->mdl_request, thd->variables.lock_wait_timeout)) /*
If a lock can't be acquired, it makes no sense to try normal insert. Therefore we just abort the statement. */ DBUG_RETURN(TRUE);
bool error= FALSE; if (delayed_get_table(thd, table_list)) error= TRUE; else if (table_list->table) { /*
Open tables used for sub-selects or in stored functions, will also cache these functions. */ if (open_and_lock_tables(thd, table_list->next_global, TRUE, 0)) { end_delayed_insert(thd); error= TRUE; } else { /*
First table was not processed by open_and_lock_tables(), we need to set updatability flag "by hand". */ if (!table_list->derived && !table_list->view) table_list->updatable= 1; // usual table
} }
/*
If a lock was acquired above, we should release it after handle_delayed_insert() has cloned the ticket. Note that acquire_lock() can succeed because the connection already has the lock. In this case the ticket will be before the mdl_savepoint and we should not release it here. */ if (!thd->mdl_context.has_lock(mdl_savepoint, table_list->mdl_request.ticket)) thd->mdl_context.release_lock(table_list->mdl_request.ticket);
/*
Reset the ticket in case we end up having to use normal insert and therefore will reopen the table and reacquire the metadata lock. */ table_list->mdl_request.ticket= NULL;
if (error || table_list->table) DBUG_RETURN(error);#endif
/*
* This is embedded library and we don't have auxiliary threads OR * a lock upgrade was requested inside delayed_get_table because - there are too many delayed insert threads OR - the table has triggers. Use a normal insert. */ table_list->lock_type= TL_WRITE; DBUG_RETURN(open_and_lock_tables(thd, table_list, TRUE, 0));}
/**
Create a new query string for removing DELAYED keyword for multi INSERT DEALAYED statement.
@param[in] thd Thread handler @param[in] buf Query string
@return 0 ok 1 error*/static intcreate_insert_stmt_from_insert_delayed(THD *thd, String *buf){ /* Make a copy of thd->query() and then remove the "DELAYED" keyword */ if (buf->append(thd->query()) || buf->replace(thd->lex->keyword_delayed_begin_offset, thd->lex->keyword_delayed_end_offset - thd->lex->keyword_delayed_begin_offset, 0)) return 1; return 0;}
/**
INSERT statement implementation
@note Like implementations of other DDL/DML in MySQL, this function relies on the caller to close the thread tables. This is done in the end of dispatch_command().*/
bool mysql_insert(THD *thd,TABLE_LIST *table_list, List<Item> &fields, List<List_item> &values_list, List<Item> &update_fields, List<Item> &update_values, enum_duplicates duplic, bool ignore){ int error, res; bool transactional_table, joins_freed= FALSE; bool changed; bool was_insert_delayed= (table_list->lock_type == TL_WRITE_DELAYED); uint value_count; ulong counter = 1; ulonglong id; COPY_INFO info; TABLE *table= 0; List_iterator_fast<List_item> its(values_list); List_item *values; Name_resolution_context *context; Name_resolution_context_state ctx_state;#ifndef EMBEDDED_LIBRARY
char *query= thd->query(); /*
log_on is about delayed inserts only. By default, both logs are enabled (this won't cause problems if the server runs without --log-bin). */ bool log_on= ((thd->variables.option_bits & OPTION_BIN_LOG) || (!(thd->security_ctx->master_access & SUPER_ACL)));#endif
thr_lock_type lock_type; Item *unused_conds= 0; DBUG_ENTER("mysql_insert");
/*
Upgrade lock type if the requested lock is incompatible with the current connection mode or table operation. */ upgrade_lock_type(thd, &table_list->lock_type, duplic, values_list.elements > 1);
/*
We can't write-delayed into a table locked with LOCK TABLES: this will lead to a deadlock, since the delayed thread will never be able to get a lock on the table. QQQ: why not upgrade the lock here instead? */ if (table_list->lock_type == TL_WRITE_DELAYED && thd->locked_tables_mode && find_locked_table(thd->open_tables, table_list->db, table_list->table_name)) { my_error(ER_DELAYED_INSERT_TABLE_LOCKED, MYF(0), table_list->table_name); DBUG_RETURN(TRUE); }
if (table_list->lock_type == TL_WRITE_DELAYED) { if (open_and_lock_for_insert_delayed(thd, table_list)) DBUG_RETURN(TRUE); } else { if (open_and_lock_tables(thd, table_list, TRUE, 0)) DBUG_RETURN(TRUE); } lock_type= table_list->lock_type;
thd_proc_info(thd, "init"); thd->used_tables=0; values= its++; value_count= values->elements;
if (mysql_prepare_insert(thd, table_list, table, fields, values, update_fields, update_values, duplic, &unused_conds, FALSE, (fields.elements || !value_count || table_list->view != 0), !ignore && (thd->variables.sql_mode & (MODE_STRICT_TRANS_TABLES | MODE_STRICT_ALL_TABLES)))) goto abort;
/* mysql_prepare_insert set table_list->table if it was not set */ table= table_list->table;
context= &thd->lex->select_lex.context; /*
These three asserts test the hypothesis that the resetting of the name resolution context below is not necessary at all since the list of local tables for INSERT always consists of one table. */ DBUG_ASSERT(!table_list->next_local); DBUG_ASSERT(!context->table_list->next_local); DBUG_ASSERT(!context->first_name_resolution_table->next_name_resolution_table);
/* Save the state of the current name resolution context. */ ctx_state.save_state(context, table_list);
/*
Perform name resolution only in the first table - 'table_list', which is the table that is inserted into. */ table_list->next_local= 0; context->resolve_in_table_list_only(table_list);
while ((values= its++)) { counter++; if (values->elements != value_count) { my_error(ER_WRONG_VALUE_COUNT_ON_ROW, MYF(0), counter); goto abort; } if (setup_fields(thd, 0, *values, MARK_COLUMNS_READ, 0, 0)) goto abort; } its.rewind (); /* Restore the current context. */ ctx_state.restore_state(context, table_list);
/*
Fill in the given fields and dump it to the table file */ bzero((char*) &info,sizeof(info)); info.ignore= ignore; info.handle_duplicates=duplic; info.update_fields= &update_fields; info.update_values= &update_values; info.view= (table_list->view ? table_list : 0);
/*
Count warnings for all inserts. For single line insert, generate an error if try to set a NOT NULL field to NULL. */ thd->count_cuted_fields= ((values_list.elements == 1 && !ignore) ? CHECK_FIELD_ERROR_FOR_NULL : CHECK_FIELD_WARN); thd->cuted_fields = 0L; table->next_number_field=table->found_next_number_field;
#ifdef HAVE_REPLICATION
if (thd->slave_thread && (info.handle_duplicates == DUP_UPDATE) && (table->next_number_field != NULL) && rpl_master_has_bug(&active_mi->rli, 24432, TRUE, NULL, NULL)) goto abort;#endif
error=0; thd_proc_info(thd, "update"); if (duplic == DUP_REPLACE && (!table->triggers || !table->triggers->has_delete_triggers())) table->file->extra(HA_EXTRA_WRITE_CAN_REPLACE); if (duplic == DUP_UPDATE) table->file->extra(HA_EXTRA_INSERT_WITH_UPDATE); /*
let's *try* to start bulk inserts. It won't necessary start them as values_list.elements should be greater than some - handler dependent - threshold. We should not start bulk inserts if this statement uses functions or invokes triggers since they may access to the same table and therefore should not see its inconsistent state created by this optimization. So we call start_bulk_insert to perform nesessary checks on values_list.elements, and - if nothing else - to initialize the code to make the call of end_bulk_insert() below safe. */#ifndef EMBEDDED_LIBRARY
if (lock_type != TL_WRITE_DELAYED)#endif /* EMBEDDED_LIBRARY */
{ if (duplic != DUP_ERROR || ignore) table->file->extra(HA_EXTRA_IGNORE_DUP_KEY); /**
This is a simple check for the case when the table has a trigger that reads from it, or when the statement invokes a stored function that reads from the table being inserted to. Engines can't handle a bulk insert in parallel with a read form the same table in the same connection. */ if (thd->locked_tables_mode <= LTM_LOCK_TABLES) table->file->ha_start_bulk_insert(values_list.elements); }
thd->abort_on_warning= (!ignore && (thd->variables.sql_mode & (MODE_STRICT_TRANS_TABLES | MODE_STRICT_ALL_TABLES)));
prepare_triggers_for_insert_stmt(table);
if (table_list->prepare_where(thd, 0, TRUE) || table_list->prepare_check_option(thd)) error= 1;
while ((values= its++)) { if (fields.elements || !value_count) { restore_record(table,s->default_values); // Get empty record
if (fill_record_n_invoke_before_triggers(thd, fields, *values, 0, table->triggers, TRG_EVENT_INSERT)) { if (values_list.elements != 1 && ! thd->is_error()) { info.records++; continue; } /*
TODO: set thd->abort_on_warning if values_list.elements == 1 and check that all items return warning in case of problem with storing field. */ error=1; break; } } else { if (thd->used_tables) // Column used in values()
restore_record(table,s->default_values); // Get empty record
else { TABLE_SHARE *share= table->s;
/*
Fix delete marker. No need to restore rest of record since it will be overwritten by fill_record() anyway (and fill_record() does not use default values in this case). */ table->record[0][0]= share->default_values[0];
/* Fix undefined null_bits. */ if (share->null_bytes > 1 && share->last_null_bit_pos) { table->record[0][share->null_bytes - 1]= share->default_values[share->null_bytes - 1]; } } if (fill_record_n_invoke_before_triggers(thd, table->field, *values, 0, table->triggers, TRG_EVENT_INSERT)) { if (values_list.elements != 1 && ! thd->is_error()) { info.records++; continue; } error=1; break; } }
if ((res= table_list->view_check_option(thd, (values_list.elements == 1 ? 0 : ignore))) == VIEW_CHECK_SKIP) continue; else if (res == VIEW_CHECK_ERROR) { error= 1; break; }#ifndef EMBEDDED_LIBRARY
if (lock_type == TL_WRITE_DELAYED) { LEX_STRING const st_query = { query, thd->query_length() }; error=write_delayed(thd, table, duplic, st_query, ignore, log_on); query=0; } else#endif
error=write_record(thd, table ,&info); if (error) break; thd->warning_info->inc_current_row_for_warning(); }
free_underlaid_joins(thd, &thd->lex->select_lex); joins_freed= TRUE;
/*
Now all rows are inserted. Time to update logs and sends response to user */#ifndef EMBEDDED_LIBRARY
if (lock_type == TL_WRITE_DELAYED) { if (!error) { info.copied=values_list.elements; end_delayed_insert(thd); } } else#endif
{ /*
Do not do this release if this is a delayed insert, it would steal auto_inc values from the delayed_insert thread as they share TABLE. */ table->file->ha_release_auto_increment(); if (thd->locked_tables_mode <= LTM_LOCK_TABLES && table->file->ha_end_bulk_insert() && !error) { table->file->print_error(my_errno,MYF(0)); error=1; } if (duplic != DUP_ERROR || ignore) table->file->extra(HA_EXTRA_NO_IGNORE_DUP_KEY);
transactional_table= table->file->has_transactions();
if ((changed= (info.copied || info.deleted || info.updated))) { /*
Invalidate the table in the query cache if something changed. For the transactional algorithm to work the invalidation must be before binlog writing and ha_autocommit_or_rollback */ query_cache_invalidate3(thd, table_list, 1); }
if (thd->transaction.stmt.modified_non_trans_table) thd->transaction.all.modified_non_trans_table= TRUE;
if ((changed && error <= 0) || thd->transaction.stmt.modified_non_trans_table || was_insert_delayed) { if (mysql_bin_log.is_open()) { int errcode= 0; if (error <= 0) { /*
[Guilhem wrote] Temporary errors may have filled thd->net.last_error/errno. For example if there has been a disk full error when writing the row, and it was MyISAM, then thd->net.last_error/errno will be set to "disk full"... and the mysql_file_pwrite() will wait until free space appears, and so when it finishes then the write_row() was entirely successful */ /* todo: consider removing */ thd->clear_error(); } else errcode= query_error_code(thd, thd->killed == THD::NOT_KILLED); /* bug#22725:
A query which per-row-loop can not be interrupted with KILLED, like INSERT, and that does not invoke stored routines can be binlogged with neglecting the KILLED error. If there was no error (error == zero) until after the end of inserting loop the KILLED flag that appeared later can be disregarded since previously possible invocation of stored routines did not result in any error due to the KILLED. In such case the flag is ignored for constructing binlog event. */ DBUG_ASSERT(thd->killed != THD::KILL_BAD_DATA || error > 0); if (was_insert_delayed && table_list->lock_type == TL_WRITE) { /* Binlog multi INSERT DELAYED as INSERT without DELAYED. */ String log_query; if (create_insert_stmt_from_insert_delayed(thd, &log_query)) { sql_print_error("Event Error: An error occurred while creating query string" "for INSERT DELAYED stmt, before writing it into binary log.");
error= 1; } else if (thd->binlog_query(THD::ROW_QUERY_TYPE, log_query.c_ptr(), log_query.length(), transactional_table, FALSE, FALSE, errcode)) error= 1; } else if (thd->binlog_query(THD::ROW_QUERY_TYPE, thd->query(), thd->query_length(), transactional_table, FALSE, FALSE, errcode)) error= 1; } } DBUG_ASSERT(transactional_table || !changed || thd->transaction.stmt.modified_non_trans_table); } thd_proc_info(thd, "end"); /*
We'll report to the client this id: - if the table contains an autoincrement column and we successfully inserted an autogenerated value, the autogenerated value. - if the table contains no autoincrement column and LAST_INSERT_ID(X) was called, X. - if the table contains an autoincrement column, and some rows were inserted, the id of the last "inserted" row (if IGNORE, that value may not have been really inserted but ignored). */ id= (thd->first_successful_insert_id_in_cur_stmt > 0) ? thd->first_successful_insert_id_in_cur_stmt : (thd->arg_of_last_insert_id_function ? thd->first_successful_insert_id_in_prev_stmt : ((table->next_number_field && info.copied) ? table->next_number_field->val_int() : 0)); table->next_number_field=0; thd->count_cuted_fields= CHECK_FIELD_IGNORE; table->auto_increment_field_not_null= FALSE; if (duplic == DUP_REPLACE && (!table->triggers || !table->triggers->has_delete_triggers())) table->file->extra(HA_EXTRA_WRITE_CANNOT_REPLACE);
if (error) goto abort; if (values_list.elements == 1 && (!(thd->variables.option_bits & OPTION_WARNINGS) || !thd->cuted_fields)) { my_ok(thd, info.copied + info.deleted + ((thd->client_capabilities & CLIENT_FOUND_ROWS) ? info.touched : info.updated), id); } else { char buff[160]; ha_rows updated=((thd->client_capabilities & CLIENT_FOUND_ROWS) ? info.touched : info.updated); if (ignore) sprintf(buff, ER(ER_INSERT_INFO), (ulong) info.records, (lock_type == TL_WRITE_DELAYED) ? (ulong) 0 : (ulong) (info.records - info.copied), (ulong) thd->warning_info->statement_warn_count()); else sprintf(buff, ER(ER_INSERT_INFO), (ulong) info.records, (ulong) (info.deleted + updated), (ulong) thd->warning_info->statement_warn_count()); ::my_ok(thd, info.copied + info.deleted + updated, id, buff); } thd->abort_on_warning= 0; DBUG_RETURN(FALSE);
abort:#ifndef EMBEDDED_LIBRARY
if (lock_type == TL_WRITE_DELAYED) end_delayed_insert(thd);#endif
if (table != NULL) table->file->ha_release_auto_increment(); if (!joins_freed) free_underlaid_joins(thd, &thd->lex->select_lex); thd->abort_on_warning= 0; DBUG_RETURN(TRUE);}
/*
Additional check for insertability for VIEW
SYNOPSIS check_view_insertability() thd - thread handler view - reference on VIEW
IMPLEMENTATION A view is insertable if the folloings are true: - All columns in the view are columns from a table - All not used columns in table have a default values - All field in view are unique (not referring to the same column)
RETURN FALSE - OK view->contain_auto_increment is 1 if and only if the view contains an auto_increment field
TRUE - can't be used for insert*/
static bool check_view_insertability(THD * thd, TABLE_LIST *view){ uint num= view->view->select_lex.item_list.elements; TABLE *table= view->table; Field_translator *trans_start= view->field_translation, *trans_end= trans_start + num; Field_translator *trans; uint used_fields_buff_size= bitmap_buffer_size(table->s->fields); uint32 *used_fields_buff= (uint32*)thd->alloc(used_fields_buff_size); MY_BITMAP used_fields; enum_mark_columns save_mark_used_columns= thd->mark_used_columns; DBUG_ENTER("check_key_in_view");
if (!used_fields_buff) DBUG_RETURN(TRUE); // EOM
DBUG_ASSERT(view->table != 0 && view->field_translation != 0);
(void) bitmap_init(&used_fields, used_fields_buff, table->s->fields, 0); bitmap_clear_all(&used_fields);
view->contain_auto_increment= 0; /*
we must not set query_id for fields as they're not really used in this context */ thd->mark_used_columns= MARK_COLUMNS_NONE; /* check simplicity and prepare unique test of view */ for (trans= trans_start; trans != trans_end; trans++) { if (!trans->item->fixed && trans->item->fix_fields(thd, &trans->item)) { thd->mark_used_columns= save_mark_used_columns; DBUG_RETURN(TRUE); } Item_field *field; /* simple SELECT list entry (field without expression) */ if (!(field= trans->item->filed_for_view_update())) { thd->mark_used_columns= save_mark_used_columns; DBUG_RETURN(TRUE); } if (field->field->unireg_check == Field::NEXT_NUMBER) view->contain_auto_increment= 1; /* prepare unique test */ /*
remove collation (or other transparent for update function) if we have it */ trans->item= field; } thd->mark_used_columns= save_mark_used_columns; /* unique test */ for (trans= trans_start; trans != trans_end; trans++) { /* Thanks to test above, we know that all columns are of type Item_field */ Item_field *field= (Item_field *)trans->item; /* check fields belong to table in which we are inserting */ if (field->field->table == table && bitmap_fast_test_and_set(&used_fields, field->field->field_index)) DBUG_RETURN(TRUE); }
DBUG_RETURN(FALSE);}
/*
Check if table can be updated
SYNOPSIS mysql_prepare_insert_check_table() thd Thread handle table_list Table list fields List of fields to be updated where Pointer to where clause select_insert Check is making for SELECT ... INSERT
RETURN FALSE ok TRUE ERROR*/
static bool mysql_prepare_insert_check_table(THD *thd, TABLE_LIST *table_list, List<Item> &fields, bool select_insert){ bool insert_into_view= (table_list->view != 0); DBUG_ENTER("mysql_prepare_insert_check_table");
/*
first table in list is the one we'll INSERT into, requires INSERT_ACL. all others require SELECT_ACL only. the ACL requirement below is for new leaves only anyway (view-constituents), so check for SELECT rather than INSERT. */
if (setup_tables_and_check_access(thd, &thd->lex->select_lex.context, &thd->lex->select_lex.top_join_list, table_list, &thd->lex->select_lex.leaf_tables, select_insert, INSERT_ACL, SELECT_ACL)) DBUG_RETURN(TRUE);
if (insert_into_view && !fields.elements) { thd->lex->empty_field_list_on_rset= 1; if (!table_list->table) { my_error(ER_VIEW_NO_INSERT_FIELD_LIST, MYF(0), table_list->view_db.str, table_list->view_name.str); DBUG_RETURN(TRUE); } DBUG_RETURN(insert_view_fields(thd, &fields, table_list)); }
DBUG_RETURN(FALSE);}
/*
Get extra info for tables we insert into
@param table table(TABLE object) we insert into, might be NULL in case of view @param table(TABLE_LIST object) or view we insert into*/
static void prepare_for_positional_update(TABLE *table, TABLE_LIST *tables){ if (table) { if(table->reginfo.lock_type != TL_WRITE_DELAYED) table->prepare_for_position(); return; }
DBUG_ASSERT(tables->view); List_iterator<TABLE_LIST> it(*tables->view_tables); TABLE_LIST *tbl; while ((tbl= it++)) prepare_for_positional_update(tbl->table, tbl);
return;}
/*
Prepare items in INSERT statement
SYNOPSIS mysql_prepare_insert() thd Thread handler table_list Global/local table list table Table to insert into (can be NULL if table should be taken from table_list->table) where Where clause (for insert ... select) select_insert TRUE if INSERT ... SELECT statement check_fields TRUE if need to check that all INSERT fields are given values. abort_on_warning whether to report if some INSERT field is not assigned as an error (TRUE) or as a warning (FALSE).
TODO (in far future) In cases of: INSERT INTO t1 SELECT a, sum(a) as sum1 from t2 GROUP BY a ON DUPLICATE KEY ... we should be able to refer to sum1 in the ON DUPLICATE KEY part
WARNING You MUST set table->insert_values to 0 after calling this function before releasing the table object. RETURN VALUE FALSE OK TRUE error*/
bool mysql_prepare_insert(THD *thd, TABLE_LIST *table_list, TABLE *table, List<Item> &fields, List_item *values, List<Item> &update_fields, List<Item> &update_values, enum_duplicates duplic, COND **where, bool select_insert, bool check_fields, bool abort_on_warning){ SELECT_LEX *select_lex= &thd->lex->select_lex; Name_resolution_context *context= &select_lex->context; Name_resolution_context_state ctx_state; bool insert_into_view= (table_list->view != 0); bool res= 0; table_map map= 0; DBUG_ENTER("mysql_prepare_insert"); DBUG_PRINT("enter", ("table_list 0x%lx, table 0x%lx, view %d", (ulong)table_list, (ulong)table, (int)insert_into_view)); /* INSERT should have a SELECT or VALUES clause */ DBUG_ASSERT (!select_insert || !values);
/*
For subqueries in VALUES() we should not see the table in which we are inserting (for INSERT ... SELECT this is done by changing table_list, because INSERT ... SELECT share SELECT_LEX it with SELECT. */ if (!select_insert) { for (SELECT_LEX_UNIT *un= select_lex->first_inner_unit(); un; un= un->next_unit()) { for (SELECT_LEX *sl= un->first_select(); sl; sl= sl->next_select()) { sl->context.outer_context= 0; } } }
if (duplic == DUP_UPDATE) { /* it should be allocated before Item::fix_fields() */ if (table_list->set_insert_values(thd->mem_root)) DBUG_RETURN(TRUE); }
if (mysql_prepare_insert_check_table(thd, table_list, fields, select_insert)) DBUG_RETURN(TRUE);
/* Prepare the fields in the statement. */ if (values) { /* if we have INSERT ... VALUES () we cannot have a GROUP BY clause */ DBUG_ASSERT (!select_lex->group_list.elements);
/* Save the state of the current name resolution context. */ ctx_state.save_state(context, table_list);
/*
Perform name resolution only in the first table - 'table_list', which is the table that is inserted into. */ table_list->next_local= 0; context->resolve_in_table_list_only(table_list);
res= (setup_fields(thd, 0, *values, MARK_COLUMNS_READ, 0, 0) || check_insert_fields(thd, context->table_list, fields, *values, !insert_into_view, 0, &map));
if (!res && check_fields) { bool saved_abort_on_warning= thd->abort_on_warning; thd->abort_on_warning= abort_on_warning; res= check_that_all_fields_are_given_values(thd, table ? table : context->table_list->table, context->table_list); thd->abort_on_warning= saved_abort_on_warning; }
if (!res) res= setup_fields(thd, 0, update_values, MARK_COLUMNS_READ, 0, 0);
if (!res && duplic == DUP_UPDATE) { select_lex->no_wrap_view_item= TRUE; res= check_update_fields(thd, context->table_list, update_fields, update_values, &map); select_lex->no_wrap_view_item= FALSE; }
/* Restore the current context. */ ctx_state.restore_state(context, table_list); }
if (res) DBUG_RETURN(res);
if (!table) table= table_list->table;
if (!select_insert) { Item *fake_conds= 0; TABLE_LIST *duplicate; if ((duplicate= unique_table(thd, table_list, table_list->next_global, 1))) { update_non_unique_table_error(table_list, "INSERT", duplicate); DBUG_RETURN(TRUE); } select_lex->fix_prepare_information(thd, &fake_conds, &fake_conds); select_lex->first_execution= 0; } /*
Only call prepare_for_posistion() if we are not performing a DELAYED operation. It will instead be executed by delayed insert thread. */ if (duplic == DUP_UPDATE || duplic == DUP_REPLACE) prepare_for_positional_update(table, table_list); DBUG_RETURN(FALSE);}
/* Check if there is more uniq keys after field */
static int last_uniq_key(TABLE *table,uint keynr){ /*
When an underlying storage engine informs that the unique key conflicts are not reported in the ascending order by setting the HA_DUPLICATE_KEY_NOT_IN_ORDER flag, we cannot rely on this information to determine the last key conflict. The information about the last key conflict will be used to do a replace of the new row on the conflicting row, rather than doing a delete (of old row) + insert (of new row). Hence check for this flag and disable replacing the last row by returning 0 always. Returning 0 will result in doing a delete + insert always. */ if (table->file->ha_table_flags() & HA_DUPLICATE_KEY_NOT_IN_ORDER) return 0;
while (++keynr < table->s->keys) if (table->key_info[keynr].flags & HA_NOSAME) return 0; return 1;}
/*
Write a record to table with optional deleting of conflicting records, invoke proper triggers if needed.
SYNOPSIS write_record() thd - thread context table - table to which record should be written info - COPY_INFO structure describing handling of duplicates and which is used for counting number of records inserted and deleted.
NOTE Once this record will be written to table after insert trigger will be invoked. If instead of inserting new record we will update old one then both on update triggers will work instead. Similarly both on delete triggers will be invoked if we will delete conflicting records.
Sets thd->transaction.stmt.modified_non_trans_table to TRUE if table which is updated didn't have transactions.
RETURN VALUE 0 - success non-0 - error*/
int write_record(THD *thd, TABLE *table,COPY_INFO *info){ int error, trg_error= 0; char *key=0; MY_BITMAP *save_read_set, *save_write_set; ulonglong prev_insert_id= table->file->next_insert_id; ulonglong insert_id_for_cur_row= 0; DBUG_ENTER("write_record");
info->records++; save_read_set= table->read_set; save_write_set= table->write_set;
if (info->handle_duplicates == DUP_REPLACE || info->handle_duplicates == DUP_UPDATE) { while ((error=table->file->ha_write_row(table->record[0]))) { uint key_nr; /*
If we do more than one iteration of this loop, from the second one the row will have an explicit value in the autoinc field, which was set at the first call of handler::update_auto_increment(). So we must save the autogenerated value to avoid thd->insert_id_for_cur_row to become 0. */ if (table->file->insert_id_for_cur_row > 0) insert_id_for_cur_row= table->file->insert_id_for_cur_row; else table->file->insert_id_for_cur_row= insert_id_for_cur_row; bool is_duplicate_key_error; if (table->file->is_fatal_error(error, HA_CHECK_DUP)) goto err; is_duplicate_key_error= table->file->is_fatal_error(error, 0); if (!is_duplicate_key_error) { /*
We come here when we had an ignorable error which is not a duplicate key error. In this we ignore error if ignore flag is set, otherwise report error as usual. We will not do any duplicate key processing. */ if (info->ignore) goto ok_or_after_trg_err; /* Ignoring a not fatal error, return 0 */ goto err; } if ((int) (key_nr = table->file->get_dup_key(error)) < 0) { error= HA_ERR_FOUND_DUPP_KEY; /* Database can't find key */ goto err; } /* Read all columns for the row we are going to replace */ table->use_all_columns(); /*
Don't allow REPLACE to replace a row when a auto_increment column was used. This ensures that we don't get a problem when the whole range of the key has been used. */ if (info->handle_duplicates == DUP_REPLACE && table->next_number_field && key_nr == table->s->next_number_index && (insert_id_for_cur_row > 0)) goto err; if (table->file->ha_table_flags() & HA_DUPLICATE_POS) { if (table->file->rnd_pos(table->record[1],table->file->dup_ref)) goto err; } else { if (table->file->extra(HA_EXTRA_FLUSH_CACHE)) /* Not needed with NISAM */ { error=my_errno; goto err; }
if (!key) { if (!(key=(char*) my_safe_alloca(table->s->max_unique_length, MAX_KEY_LENGTH))) { error=ENOMEM; goto err; } } key_copy((uchar*) key,table->record[0],table->key_info+key_nr,0); if ((error=(table->file->index_read_idx_map(table->record[1],key_nr, (uchar*) key, HA_WHOLE_KEY, HA_READ_KEY_EXACT)))) goto err; } if (info->handle_duplicates == DUP_UPDATE) { int res= 0; /*
We don't check for other UNIQUE keys - the first row that matches, is updated. If update causes a conflict again, an error is returned */ DBUG_ASSERT(table->insert_values != NULL); store_record(table,insert_values); restore_record(table,record[1]); DBUG_ASSERT(info->update_fields->elements == info->update_values->elements); if (fill_record_n_invoke_before_triggers(thd, *info->update_fields, *info->update_values, info->ignore, table->triggers, TRG_EVENT_UPDATE)) goto before_trg_err;
/* CHECK OPTION for VIEW ... ON DUPLICATE KEY UPDATE ... */ if (info->view && (res= info->view->view_check_option(current_thd, info->ignore)) == VIEW_CHECK_SKIP) goto ok_or_after_trg_err; if (res == VIEW_CHECK_ERROR) goto before_trg_err;
table->file->restore_auto_increment(prev_insert_id); if (table->next_number_field) table->file->adjust_next_insert_id_after_explicit_value( table->next_number_field->val_int()); info->touched++; if (!records_are_comparable(table) || compare_records(table)) { if ((error=table->file->ha_update_row(table->record[1], table->record[0])) && error != HA_ERR_RECORD_IS_THE_SAME) { if (info->ignore && !table->file->is_fatal_error(error, HA_CHECK_DUP_KEY)) { goto ok_or_after_trg_err; } goto err; }
if (error != HA_ERR_RECORD_IS_THE_SAME) info->updated++; else error= 0; /*
If ON DUP KEY UPDATE updates a row instead of inserting one, it's like a regular UPDATE statement: it should not affect the value of a next SELECT LAST_INSERT_ID() or mysql_insert_id(). Except if LAST_INSERT_ID(#) was in the INSERT query, which is handled separately by THD::arg_of_last_insert_id_function. */ insert_id_for_cur_row= table->file->insert_id_for_cur_row= 0; trg_error= (table->triggers && table->triggers->process_triggers(thd, TRG_EVENT_UPDATE, TRG_ACTION_AFTER, TRUE)); info->copied++; }
if (table->next_number_field) table->file->adjust_next_insert_id_after_explicit_value( table->next_number_field->val_int()); info->touched++;
goto ok_or_after_trg_err; } else /* DUP_REPLACE */ { /*
The manual defines the REPLACE semantics that it is either an INSERT or DELETE(s) + INSERT; FOREIGN KEY checks in InnoDB do not function in the defined way if we allow MySQL to convert the latter operation internally to an UPDATE. We also should not perform this conversion if we have timestamp field with ON UPDATE which is different from DEFAULT. Another case when conversion should not be performed is when we have ON DELETE trigger on table so user may notice that we cheat here. Note that it is ok to do such conversion for tables which have ON UPDATE but have no ON DELETE triggers, we just should not expose this fact to users by invoking ON UPDATE triggers. */ if (last_uniq_key(table,key_nr) && !table->file->referenced_by_foreign_key() && (table->timestamp_field_type == TIMESTAMP_NO_AUTO_SET || table->timestamp_field_type == TIMESTAMP_AUTO_SET_ON_BOTH) && (!table->triggers || !table->triggers->has_delete_triggers())) { if ((error=table->file->ha_update_row(table->record[1], table->record[0])) && error != HA_ERR_RECORD_IS_THE_SAME) goto err; if (error != HA_ERR_RECORD_IS_THE_SAME) info->deleted++; else error= 0; thd->record_first_successful_insert_id_in_cur_stmt(table->file->insert_id_for_cur_row); /*
Since we pretend that we have done insert we should call its after triggers. */ goto after_trg_n_copied_inc; } else { if (table->triggers && table->triggers->process_triggers(thd, TRG_EVENT_DELETE, TRG_ACTION_BEFORE, TRUE)) goto before_trg_err; if ((error=table->file->ha_delete_row(table->record[1]))) goto err; info->deleted++; if (!table->file->has_transactions()) thd->transaction.stmt.modified_non_trans_table= TRUE; if (table->triggers && table->triggers->process_triggers(thd, TRG_EVENT_DELETE, TRG_ACTION_AFTER, TRUE)) { trg_error= 1; goto ok_or_after_trg_err; } /* Let us attempt do write_row() once more */ } } } /*
If more than one iteration of the above while loop is done, from the second one the row being inserted will have an explicit value in the autoinc field, which was set at the first call of handler::update_auto_increment(). This value is saved to avoid thd->insert_id_for_cur_row becoming 0. Use this saved autoinc value. */ if (table->file->insert_id_for_cur_row == 0) table->file->insert_id_for_cur_row= insert_id_for_cur_row; thd->record_first_successful_insert_id_in_cur_stmt(table->file->insert_id_for_cur_row); /*
Restore column maps if they where replaced during an duplicate key problem. */ if (table->read_set != save_read_set || table->write_set != save_write_set) table->column_bitmaps_set(save_read_set, save_write_set); } else if ((error=table->file->ha_write_row(table->record[0]))) { if (!info->ignore || table->file->is_fatal_error(error, HA_CHECK_DUP)) goto err; table->file->restore_auto_increment(prev_insert_id); goto ok_or_after_trg_err; }
after_trg_n_copied_inc: info->copied++; thd->record_first_successful_insert_id_in_cur_stmt(table->file->insert_id_for_cur_row); trg_error= (table->triggers && table->triggers->process_triggers(thd, TRG_EVENT_INSERT, TRG_ACTION_AFTER, TRUE));
ok_or_after_trg_err: if (key) my_safe_afree(key,table->s->max_unique_length,MAX_KEY_LENGTH); if (!table->file->has_transactions()) thd->transaction.stmt.modified_non_trans_table= TRUE; DBUG_RETURN(trg_error);
err: info->last_errno= error; /* current_select is NULL if this is a delayed insert */ if (thd->lex->current_select) thd->lex->current_select->no_error= 0; // Give error
table->file->print_error(error,MYF(0)); before_trg_err: table->file->restore_auto_increment(prev_insert_id); if (key) my_safe_afree(key, table->s->max_unique_length, MAX_KEY_LENGTH); table->column_bitmaps_set(save_read_set, save_write_set); DBUG_RETURN(1);}
/******************************************************************************
Check that all fields with arn't null_fields are used******************************************************************************/
int check_that_all_fields_are_given_values(THD *thd, TABLE *entry, TABLE_LIST *table_list){ int err= 0; MY_BITMAP *write_set= entry->write_set;
for (Field **field=entry->field ; *field ; field++) { if (!bitmap_is_set(write_set, (*field)->field_index) && ((*field)->flags & NO_DEFAULT_VALUE_FLAG) && ((*field)->real_type() != MYSQL_TYPE_ENUM)) { bool view= FALSE; if (table_list) { table_list= table_list->top_table(); view= test(table_list->view); } if (view) { push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_WARN, ER_NO_DEFAULT_FOR_VIEW_FIELD, ER(ER_NO_DEFAULT_FOR_VIEW_FIELD), table_list->view_db.str, table_list->view_name.str); } else { push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_WARN, ER_NO_DEFAULT_FOR_FIELD, ER(ER_NO_DEFAULT_FOR_FIELD), (*field)->field_name); } err= 1; } } return thd->abort_on_warning ? err : 0;}
/*****************************************************************************
Handling of delayed inserts A thread is created for each table that one uses with the DELAYED attribute.*****************************************************************************/
#ifndef EMBEDDED_LIBRARY
class delayed_row :public ilink {public: char *record; enum_duplicates dup; time_t start_time; ulong sql_mode; bool auto_increment_field_not_null; bool query_start_used, ignore, log_query; bool stmt_depends_on_first_successful_insert_id_in_prev_stmt; ulonglong first_successful_insert_id_in_prev_stmt; ulonglong forced_insert_id; ulong auto_increment_increment; ulong auto_increment_offset; timestamp_auto_set_type timestamp_field_type; LEX_STRING query; Time_zone *time_zone;
delayed_row(LEX_STRING const query_arg, enum_duplicates dup_arg, bool ignore_arg, bool log_query_arg) : record(0), dup(dup_arg), ignore(ignore_arg), log_query(log_query_arg), forced_insert_id(0), query(query_arg), time_zone(0) {} ~delayed_row() { my_free(query.str); my_free(record); }};
/**
Delayed_insert - context of a thread responsible for delayed insert into one table. When processing delayed inserts, we create an own thread for every distinct table. Later on all delayed inserts directed into that table are handled by a dedicated thread.*/
class Delayed_insert :public ilink { uint locks_in_memory; thr_lock_type delayed_lock;public: THD thd; TABLE *table; mysql_mutex_t mutex; mysql_cond_t cond, cond_client; volatile uint tables_in_use,stacked_inserts; volatile bool status; /*
When the handler thread starts, it clones a metadata lock ticket for the table to be inserted. This is done to allow the deadlock detector to detect deadlocks resulting from this lock. Before this is done, the connection thread cannot safely exit without causing problems for clone_ticket(). Once handler_thread_initialized has been set, it is safe for the connection thread to exit. Access to handler_thread_initialized is protected by di->mutex. */ bool handler_thread_initialized; COPY_INFO info; I_List<delayed_row> rows; ulong group_count; TABLE_LIST table_list; // Argument
Delayed_insert() :locks_in_memory(0), table(0),tables_in_use(0),stacked_inserts(0), status(0), handler_thread_initialized(FALSE), group_count(0) { DBUG_ENTER("Delayed_insert constructor"); thd.security_ctx->user=(char*) delayed_user; thd.security_ctx->host=(char*) my_localhost; strmake(thd.security_ctx->priv_user, thd.security_ctx->user, USERNAME_LENGTH); thd.current_tablenr=0; thd.command=COM_DELAYED_INSERT; thd.lex->current_select= 0; // for my_message_sql
thd.lex->sql_command= SQLCOM_INSERT; // For innodb::store_lock()
/*
Statement-based replication of INSERT DELAYED has problems with RAND() and user variables, so in mixed mode we go to row-based. For normal commands, the unsafe flag is set at parse time. However, since the flag is a member of the THD object, of which the delayed_insert thread has its own copy, we must set the statement to unsafe here and explicitly set row logging mode.
@todo set_current_stmt_binlog_format_row_if_mixed should not be called by anything else than thd->decide_logging_format(). When we call set_current_blah here, none of the checks in decide_logging_format is made. We should probably call thd->decide_logging_format() directly instead. /Sven */ thd.lex->set_stmt_unsafe(LEX::BINLOG_STMT_UNSAFE_INSERT_DELAYED); thd.set_current_stmt_binlog_format_row_if_mixed(); /*
Prevent changes to global.lock_wait_timeout from affecting delayed insert threads as any timeouts in delayed inserts are not communicated to the client. */ thd.variables.lock_wait_timeout= LONG_TIMEOUT;
bzero((char*) &thd.net, sizeof(thd.net)); // Safety
bzero((char*) &table_list, sizeof(table_list)); // Safety
thd.system_thread= SYSTEM_THREAD_DELAYED_INSERT; thd.security_ctx->host_or_ip= ""; bzero((char*) &info,sizeof(info)); mysql_mutex_init(key_delayed_insert_mutex, &mutex, MY_MUTEX_INIT_FAST); mysql_cond_init(key_delayed_insert_cond, &cond, NULL); mysql_cond_init(key_delayed_insert_cond_client, &cond_client, NULL); mysql_mutex_lock(&LOCK_thread_count); delayed_insert_threads++; delayed_lock= global_system_variables.low_priority_updates ? TL_WRITE_LOW_PRIORITY : TL_WRITE; mysql_mutex_unlock(&LOCK_thread_count); DBUG_VOID_RETURN; } ~Delayed_insert() { /* The following is not really needed, but just for safety */ delayed_row *row; while ((row=rows.get())) delete row; if (table) { close_thread_tables(&thd); thd.mdl_context.release_transactional_locks(); } mysql_mutex_lock(&LOCK_thread_count); mysql_mutex_destroy(&mutex); mysql_cond_destroy(&cond); mysql_cond_destroy(&cond_client); thd.unlink(); // Must be unlinked under lock
my_free(thd.query()); thd.security_ctx->user= thd.security_ctx->host=0; thread_count--; delayed_insert_threads--; mysql_mutex_unlock(&LOCK_thread_count); mysql_cond_broadcast(&COND_thread_count); /* Tell main we are ready */ }
/* The following is for checking when we can delete ourselves */ inline void lock() { locks_in_memory++; // Assume LOCK_delay_insert
} void unlock() { mysql_mutex_lock(&LOCK_delayed_insert); if (!--locks_in_memory) { mysql_mutex_lock(&mutex); if (thd.killed && ! stacked_inserts && ! tables_in_use) { mysql_cond_signal(&cond); status=1; } mysql_mutex_unlock(&mutex); } mysql_mutex_unlock(&LOCK_delayed_insert); } inline uint lock_count() { return locks_in_memory; }
TABLE* get_local_table(THD* client_thd); bool open_and_lock_table(); bool handle_inserts(void);};
I_List<Delayed_insert> delayed_threads;
/**
Return an instance of delayed insert thread that can handle inserts into a given table, if it exists. Otherwise return NULL.*/
staticDelayed_insert *find_handler(THD *thd, TABLE_LIST *table_list){ thd_proc_info(thd, "waiting for delay_list"); mysql_mutex_lock(&LOCK_delayed_insert); // Protect master list
I_List_iterator<Delayed_insert> it(delayed_threads); Delayed_insert *di; while ((di= it++)) { if (!strcmp(table_list->db, di->table_list.db) && !strcmp(table_list->table_name, di->table_list.table_name)) { di->lock(); break; } } mysql_mutex_unlock(&LOCK_delayed_insert); // For unlink from list
return di;}
/**
Attempt to find or create a delayed insert thread to handle inserts into this table.
@return In case of success, table_list->table points to a local copy of the delayed table or is set to NULL, which indicates a request for lock upgrade. In case of failure, value of table_list->table is undefined. @retval TRUE - this thread ran out of resources OR - a newly created delayed insert thread ran out of resources OR - the created thread failed to open and lock the table (e.g. because it does not exist) OR - the table opened in the created thread turned out to be a view @retval FALSE - table successfully opened OR - too many delayed insert threads OR - the table has triggers and we have to fall back to a normal INSERT Two latter cases indicate a request for lock upgrade.
XXX: why do we regard INSERT DELAYED into a view as an error and do not simply perform a lock upgrade?
TODO: The approach with using two mutexes to work with the delayed thread list -- LOCK_delayed_insert and LOCK_delayed_create -- is redundant, and we only need one of them to protect the list. The reason we have two locks is that we do not want to block look-ups in the list while we're waiting for the newly created thread to open the delayed table. However, this wait itself is redundant -- we always call get_local_table later on, and there wait again until the created thread acquires a table lock.
As is redundant the concept of locks_in_memory, since we already have another counter with similar semantics - tables_in_use, both of them are devoted to counting the number of producers for a given consumer (delayed insert thread), only at different stages of producer-consumer relationship.
The 'status' variable in Delayed_insert is redundant too, since there is already di->stacked_inserts.*/
staticbool delayed_get_table(THD *thd, TABLE_LIST *table_list){ int error; Delayed_insert *di; DBUG_ENTER("delayed_get_table");
/* Must be set in the parser */ DBUG_ASSERT(table_list->db);
/* Find the thread which handles this table. */ if (!(di= find_handler(thd, table_list))) { /*
No match. Create a new thread to handle the table, but no more than max_insert_delayed_threads. */ if (delayed_insert_threads >= thd->variables.max_insert_delayed_threads) DBUG_RETURN(0); thd_proc_info(thd, "Creating delayed handler"); mysql_mutex_lock(&LOCK_delayed_create); /*
The first search above was done without LOCK_delayed_create. Another thread might have created the handler in between. Search again. */ if (! (di= find_handler(thd, table_list))) { if (!(di= new Delayed_insert())) goto end_create; mysql_mutex_lock(&LOCK_thread_count); thread_count++; mysql_mutex_unlock(&LOCK_thread_count); di->thd.set_db(table_list->db, (uint) strlen(table_list->db)); di->thd.set_query(my_strdup(table_list->table_name, MYF(MY_WME | ME_FATALERROR)), 0); if (di->thd.db == NULL || di->thd.query() == NULL) { /* The error is reported */ delete di; goto end_create; } di->table_list= *table_list; // Needed to open table
/* Replace volatile strings with local copies */ di->table_list.alias= di->table_list.table_name= di->thd.query(); di->table_list.db= di->thd.db; /* We need the ticket so that it can be cloned in handle_delayed_insert */ init_mdl_requests(&di->table_list); di->table_list.mdl_request.ticket= table_list->mdl_request.ticket;
di->lock(); mysql_mutex_lock(&di->mutex); if ((error= mysql_thread_create(key_thread_delayed_insert, &di->thd.real_id, &connection_attrib, handle_delayed_insert, (void*) di))) { DBUG_PRINT("error", ("Can't create thread to handle delayed insert (error %d)", error)); mysql_mutex_unlock(&di->mutex); di->unlock(); delete di; my_error(ER_CANT_CREATE_THREAD, MYF(ME_FATALERROR), error); goto end_create; }
/*
Wait until table is open unless the handler thread or the connection thread has been killed. Note that we in all cases must wait until the handler thread has been properly initialized before exiting. Otherwise we risk doing clone_ticket() on a ticket that is no longer valid. */ thd_proc_info(thd, "waiting for handler open"); while (!di->handler_thread_initialized || (!di->thd.killed && !di->table && !thd->killed)) { mysql_cond_wait(&di->cond_client, &di->mutex); } mysql_mutex_unlock(&di->mutex); thd_proc_info(thd, "got old table"); if (thd->killed) { di->unlock(); goto end_create; } if (di->thd.killed) { if (di->thd.is_error()) { /*
Copy the error message. Note that we don't treat fatal errors in the delayed thread as fatal errors in the main thread. If delayed thread was killed, we don't want to send "Server shutdown in progress" in the INSERT THREAD. */ if (di->thd.stmt_da->sql_errno() == ER_SERVER_SHUTDOWN) my_message(ER_QUERY_INTERRUPTED, ER(ER_QUERY_INTERRUPTED), MYF(0)); else my_message(di->thd.stmt_da->sql_errno(), di->thd.stmt_da->message(), MYF(0)); } di->unlock(); goto end_create; } mysql_mutex_lock(&LOCK_delayed_insert); delayed_threads.append(di); mysql_mutex_unlock(&LOCK_delayed_insert); } mysql_mutex_unlock(&LOCK_delayed_create); }
mysql_mutex_lock(&di->mutex); table_list->table= di->get_local_table(thd); mysql_mutex_unlock(&di->mutex); if (table_list->table) { DBUG_ASSERT(! thd->is_error()); thd->di= di; } /* Unlock the delayed insert object after its last access. */ di->unlock(); DBUG_RETURN((table_list->table == NULL));
end_create: mysql_mutex_unlock(&LOCK_delayed_create); DBUG_RETURN(thd->is_error());}
/**
As we can't let many client threads modify the same TABLE structure of the dedicated delayed insert thread, we create an own structure for each client thread. This includes a row buffer to save the column values and new fields that point to the new row buffer. The memory is allocated in the client thread and is freed automatically.
@pre This function is called from the client thread. Delayed insert thread mutex must be acquired before invoking this function.
@return Not-NULL table object on success. NULL in case of an error, which is set in client_thd.*/
TABLE *Delayed_insert::get_local_table(THD* client_thd){ my_ptrdiff_t adjust_ptrs; Field **field,**org_field, *found_next_number_field; TABLE *copy; TABLE_SHARE *share; uchar *bitmap; DBUG_ENTER("Delayed_insert::get_local_table");
/* First request insert thread to get a lock */ status=1; tables_in_use++; if (!thd.lock) // Table is not locked
{ thd_proc_info(client_thd, "waiting for handler lock"); mysql_cond_signal(&cond); // Tell handler to lock table
while (!thd.killed && !thd.lock && ! client_thd->killed) { mysql_cond_wait(&cond_client, &mutex); } thd_proc_info(client_thd, "got handler lock"); if (client_thd->killed) goto error; if (thd.killed) { /*
Copy the error message. Note that we don't treat fatal errors in the delayed thread as fatal errors in the main thread. If delayed thread was killed, we don't want to send "Server shutdown in progress" in the INSERT THREAD.
The thread could be killed with an error message if di->handle_inserts() or di->open_and_lock_table() fails. The thread could be killed without an error message if killed using mysql_notify_thread_having_shared_lock() or kill_delayed_threads_for_table(). */ if (!thd.is_error() || thd.stmt_da->sql_errno() == ER_SERVER_SHUTDOWN) my_message(ER_QUERY_INTERRUPTED, ER(ER_QUERY_INTERRUPTED), MYF(0)); else my_message(thd.stmt_da->sql_errno(), thd.stmt_da->message(), MYF(0)); goto error; } } share= table->s;
/*
Allocate memory for the TABLE object, the field pointers array, and one record buffer of reclength size. Normally a table has three record buffers of rec_buff_length size, which includes alignment bytes. Since the table copy is used for creating one record only, the other record buffers and alignment are unnecessary. */ thd_proc_info(client_thd, "allocating local table"); copy= (TABLE*) client_thd->alloc(sizeof(*copy)+ (share->fields+1)*sizeof(Field**)+ share->reclength + share->column_bitmap_size*2); if (!copy) goto error;
/* Copy the TABLE object. */ *copy= *table; /* We don't need to change the file handler here */ /* Assign the pointers for the field pointers array and the record. */ field= copy->field= (Field**) (copy + 1); bitmap= (uchar*) (field + share->fields + 1); copy->record[0]= (bitmap + share->column_bitmap_size * 2); memcpy((char*) copy->record[0], (char*) table->record[0], share->reclength); /*
Make a copy of all fields. The copied fields need to point into the copied record. This is done by copying the field objects with their old pointer values and then "move" the pointers by the distance between the original and copied records. That way we preserve the relative positions in the records. */ adjust_ptrs= PTR_BYTE_DIFF(copy->record[0], table->record[0]); found_next_number_field= table->found_next_number_field; for (org_field= table->field; *org_field; org_field++, field++) { if (!(*field= (*org_field)->new_field(client_thd->mem_root, copy, 1))) goto error; (*field)->orig_table= copy; // Remove connection
(*field)->move_field_offset(adjust_ptrs); // Point at copy->record[0]
if (*org_field == found_next_number_field) (*field)->table->found_next_number_field= *field; } *field=0;
/* Adjust timestamp */ if (table->timestamp_field) { /* Restore offset as this may have been reset in handle_inserts */ copy->timestamp_field= (Field_timestamp*) copy->field[share->timestamp_field_offset]; copy->timestamp_field->unireg_check= table->timestamp_field->unireg_check; copy->timestamp_field_type= copy->timestamp_field->get_auto_set_type(); }
/* Adjust in_use for pointing to client thread */ copy->in_use= client_thd;
/* Adjust lock_count. This table object is not part of a lock. */ copy->lock_count= 0;
/* Adjust bitmaps */ copy->def_read_set.bitmap= (my_bitmap_map*) bitmap; copy->def_write_set.bitmap= ((my_bitmap_map*) (bitmap + share->column_bitmap_size)); copy->tmp_set.bitmap= 0; // To catch errors
bzero((char*) bitmap, share->column_bitmap_size*2); copy->read_set= ©->def_read_set; copy->write_set= ©->def_write_set;
DBUG_RETURN(copy);
/* Got fatal error */ error: tables_in_use--; status=1; mysql_cond_signal(&cond); // Inform thread about abort
DBUG_RETURN(0);}
/* Put a question in queue */
staticint write_delayed(THD *thd, TABLE *table, enum_duplicates duplic, LEX_STRING query, bool ignore, bool log_on){ delayed_row *row= 0; Delayed_insert *di=thd->di; const Discrete_interval *forced_auto_inc; DBUG_ENTER("write_delayed"); DBUG_PRINT("enter", ("query = '%s' length %lu", query.str, (ulong) query.length));
thd_proc_info(thd, "waiting for handler insert"); mysql_mutex_lock(&di->mutex); while (di->stacked_inserts >= delayed_queue_size && !thd->killed) mysql_cond_wait(&di->cond_client, &di->mutex); thd_proc_info(thd, "storing row into queue");
if (thd->killed) goto err;
/*
Take a copy of the query string, if there is any. The string will be free'ed when the row is destroyed. If there is no query string, we don't do anything special. */
if (query.str) { char *str; if (!(str= my_strndup(query.str, query.length, MYF(MY_WME)))) goto err; query.str= str; } row= new delayed_row(query, duplic, ignore, log_on); if (row == NULL) { my_free(query.str); goto err; }
if (!(row->record= (char*) my_malloc(table->s->reclength, MYF(MY_WME)))) goto err; memcpy(row->record, table->record[0], table->s->reclength); row->start_time= thd->start_time; row->query_start_used= thd->query_start_used; /*
those are for the binlog: LAST_INSERT_ID() has been evaluated at this time, so record does not need it, but statement-based binlogging of the INSERT will need when the row is actually inserted. As for SET INSERT_ID, DELAYED does not honour it (BUG#20830). */ row->stmt_depends_on_first_successful_insert_id_in_prev_stmt= thd->stmt_depends_on_first_successful_insert_id_in_prev_stmt; row->first_successful_insert_id_in_prev_stmt= thd->first_successful_insert_id_in_prev_stmt; row->timestamp_field_type= table->timestamp_field_type;
/* Add session variable timezone
Time_zone object will not be freed even the thread is ended. So we can get time_zone object from thread which handling delayed statement. See the comment of my_tz_find() for detail. */ if (thd->time_zone_used) { row->time_zone = thd->variables.time_zone; } else { row->time_zone = NULL; } /* Copy session variables. */ row->auto_increment_increment= thd->variables.auto_increment_increment; row->auto_increment_offset= thd->variables.auto_increment_offset; row->sql_mode= thd->variables.sql_mode; row->auto_increment_field_not_null= table->auto_increment_field_not_null;
/* Copy the next forced auto increment value, if any. */ if ((forced_auto_inc= thd->auto_inc_intervals_forced.get_next())) { row->forced_insert_id= forced_auto_inc->minimum(); DBUG_PRINT("delayed", ("transmitting auto_inc: %lu", (ulong) row->forced_insert_id)); }
di->rows.push_back(row); di->stacked_inserts++; di->status=1; if (table->s->blob_fields) unlink_blobs(table); mysql_cond_signal(&di->cond);
thread_safe_increment(delayed_rows_in_use,&LOCK_delayed_status); mysql_mutex_unlock(&di->mutex); DBUG_RETURN(0);
err: delete row; mysql_mutex_unlock(&di->mutex); DBUG_RETURN(1);}
/**
Signal the delayed insert thread that this user connection is finished using it for this statement.*/
static void end_delayed_insert(THD *thd){ DBUG_ENTER("end_delayed_insert"); Delayed_insert *di=thd->di; mysql_mutex_lock(&di->mutex); DBUG_PRINT("info",("tables in use: %d",di->tables_in_use)); if (!--di->tables_in_use || di->thd.killed) { // Unlock table
di->status=1; mysql_cond_signal(&di->cond); } mysql_mutex_unlock(&di->mutex); DBUG_VOID_RETURN;}
/* We kill all delayed threads when doing flush-tables */
void kill_delayed_threads(void){ mysql_mutex_lock(&LOCK_delayed_insert); // For unlink from list
I_List_iterator<Delayed_insert> it(delayed_threads); Delayed_insert *di; while ((di= it++)) { di->thd.killed= THD::KILL_CONNECTION; if (di->thd.mysys_var) { mysql_mutex_lock(&di->thd.mysys_var->mutex); if (di->thd.mysys_var->current_cond) { /*
We need the following test because the main mutex may be locked in handle_delayed_insert() */ if (&di->mutex != di->thd.mysys_var->current_mutex) mysql_mutex_lock(di->thd.mysys_var->current_mutex); mysql_cond_broadcast(di->thd.mysys_var->current_cond); if (&di->mutex != di->thd.mysys_var->current_mutex) mysql_mutex_unlock(di->thd.mysys_var->current_mutex); } mysql_mutex_unlock(&di->thd.mysys_var->mutex); } } mysql_mutex_unlock(&LOCK_delayed_insert); // For unlink from list
}
/**
A strategy for the prelocking algorithm which prevents the delayed insert thread from opening tables with engines which do not support delayed inserts.
Particularly it allows to abort open_tables() as soon as we discover that we have opened a MERGE table, without acquiring metadata locks on underlying tables.*/
class Delayed_prelocking_strategy : public Prelocking_strategy{public: virtual bool handle_routine(THD *thd, Query_tables_list *prelocking_ctx, Sroutine_hash_entry *rt, sp_head *sp, bool *need_prelocking); virtual bool handle_table(THD *thd, Query_tables_list *prelocking_ctx, TABLE_LIST *table_list, bool *need_prelocking); virtual bool handle_view(THD *thd, Query_tables_list *prelocking_ctx, TABLE_LIST *table_list, bool *need_prelocking);};
bool Delayed_prelocking_strategy::handle_table(THD *thd, Query_tables_list *prelocking_ctx, TABLE_LIST *table_list, bool *need_prelocking){ DBUG_ASSERT(table_list->lock_type == TL_WRITE_DELAYED);
if (!(table_list->table->file->ha_table_flags() & HA_CAN_INSERT_DELAYED)) { my_error(ER_DELAYED_NOT_SUPPORTED, MYF(0), table_list->table_name); return TRUE; } return FALSE;}
bool Delayed_prelocking_strategy::handle_routine(THD *thd, Query_tables_list *prelocking_ctx, Sroutine_hash_entry *rt, sp_head *sp, bool *need_prelocking){ /* LEX used by the delayed insert thread has no routines. */ DBUG_ASSERT(0); return FALSE;}
bool Delayed_prelocking_strategy::handle_view(THD *thd, Query_tables_list *prelocking_ctx, TABLE_LIST *table_list, bool *need_prelocking){ /* We don't open views in the delayed insert thread. */ DBUG_ASSERT(0); return FALSE;}
/**
Open and lock table for use by delayed thread and check that this table is suitable for delayed inserts.
@retval FALSE - Success. @retval TRUE - Failure.*/
bool Delayed_insert::open_and_lock_table(){ Delayed_prelocking_strategy prelocking_strategy;
/*
Use special prelocking strategy to get ER_DELAYED_NOT_SUPPORTED error for tables with engines which don't support delayed inserts. */ if (!(table= open_n_lock_single_table(&thd, &table_list, TL_WRITE_DELAYED, MYSQL_OPEN_IGNORE_GLOBAL_READ_LOCK, &prelocking_strategy))) { thd.fatal_error(); // Abort waiting inserts
return TRUE; }
if (table->triggers) { /*
Table has triggers. This is not an error, but we do not support triggers with delayed insert. Terminate the delayed thread without an error and thus request lock upgrade. */ return TRUE; } table->copy_blobs= 1; return FALSE;}
/*
* Create a new delayed insert thread*/
pthread_handler_t handle_delayed_insert(void *arg){ Delayed_insert *di=(Delayed_insert*) arg; THD *thd= &di->thd;
pthread_detach_this_thread(); /* Add thread to THD list so that's it's visible in 'show processlist' */ mysql_mutex_lock(&LOCK_thread_count); thd->thread_id= thd->variables.pseudo_thread_id= thread_id++; thd->set_current_time(); threads.append(thd); thd->killed=abort_loop ? THD::KILL_CONNECTION : THD::NOT_KILLED; mysql_mutex_unlock(&LOCK_thread_count);
mysql_thread_set_psi_id(thd->thread_id);
/*
Wait until the client runs into mysql_cond_wait(), where we free it after the table is opened and di linked in the list. If we did not wait here, the client might detect the opened table before it is linked to the list. It would release LOCK_delayed_create and allow another thread to create another handler for the same table, since it does not find one in the list. */ mysql_mutex_lock(&di->mutex); if (my_thread_init()) { /* Can't use my_error since store_globals has not yet been called */ thd->stmt_da->set_error_status(thd, ER_OUT_OF_RESOURCES, ER(ER_OUT_OF_RESOURCES), NULL); di->handler_thread_initialized= TRUE; } else { DBUG_ENTER("handle_delayed_insert"); thd->thread_stack= (char*) &thd; if (init_thr_lock() || thd->store_globals()) { /* Can't use my_error since store_globals has perhaps failed */ thd->stmt_da->set_error_status(thd, ER_OUT_OF_RESOURCES, ER(ER_OUT_OF_RESOURCES), NULL); di->handler_thread_initialized= TRUE; thd->fatal_error(); goto err; }
thd->lex->sql_command= SQLCOM_INSERT; // For innodb::store_lock()
/*
Statement-based replication of INSERT DELAYED has problems with RAND() and user vars, so in mixed mode we go to row-based. */ thd->lex->set_stmt_unsafe(LEX::BINLOG_STMT_UNSAFE_INSERT_DELAYED); thd->set_current_stmt_binlog_format_row_if_mixed();
/*
Clone the ticket representing the lock on the target table for the insert and add it to the list of granted metadata locks held by the handler thread. This is safe since the handler thread is not holding nor waiting on any metadata locks. */ if (thd->mdl_context.clone_ticket(&di->table_list.mdl_request)) { di->handler_thread_initialized= TRUE; goto err; }
/*
Now that the ticket has been cloned, it is safe for the connection thread to exit. */ di->handler_thread_initialized= TRUE; di->table_list.mdl_request.ticket= NULL;
if (di->open_and_lock_table()) goto err;
/* Tell client that the thread is initialized */ mysql_cond_signal(&di->cond_client);
/* Now wait until we get an insert or lock to handle */ /* We will not abort as long as a client thread uses this thread */
for (;;) { if (thd->killed) { uint lock_count; /*
Remove this from delay insert list so that no one can request a table from this */ mysql_mutex_unlock(&di->mutex); mysql_mutex_lock(&LOCK_delayed_insert); di->unlink(); lock_count=di->lock_count(); mysql_mutex_unlock(&LOCK_delayed_insert); mysql_mutex_lock(&di->mutex); if (!lock_count && !di->tables_in_use && !di->stacked_inserts) break; // Time to die
}
/* Shouldn't wait if killed or an insert is waiting. */ if (!thd->killed && !di->status && !di->stacked_inserts) { struct timespec abstime; set_timespec(abstime, delayed_insert_timeout);
/* Information for pthread_kill */ di->thd.mysys_var->current_mutex= &di->mutex; di->thd.mysys_var->current_cond= &di->cond; thd_proc_info(&(di->thd), "Waiting for INSERT");
DBUG_PRINT("info",("Waiting for someone to insert rows")); while (!thd->killed && !di->status) { int error; mysql_audit_release(thd);#if defined(HAVE_BROKEN_COND_TIMEDWAIT)
error= mysql_cond_wait(&di->cond, &di->mutex);#else
error= mysql_cond_timedwait(&di->cond, &di->mutex, &abstime);#ifdef EXTRA_DEBUG
if (error && error != EINTR && error != ETIMEDOUT) { fprintf(stderr, "Got error %d from mysql_cond_timedwait\n", error); DBUG_PRINT("error", ("Got error %d from mysql_cond_timedwait", error)); }#endif
#endif
if (error == ETIMEDOUT || error == ETIME) thd->killed= THD::KILL_CONNECTION; } /* We can't lock di->mutex and mysys_var->mutex at the same time */ mysql_mutex_unlock(&di->mutex); mysql_mutex_lock(&di->thd.mysys_var->mutex); di->thd.mysys_var->current_mutex= 0; di->thd.mysys_var->current_cond= 0; mysql_mutex_unlock(&di->thd.mysys_var->mutex); mysql_mutex_lock(&di->mutex); } thd_proc_info(&(di->thd), 0);
if (di->tables_in_use && ! thd->lock && !thd->killed) { /*
Request for new delayed insert. Lock the table, but avoid to be blocked by a global read lock. If we got here while a global read lock exists, then one or more inserts started before the lock was requested. These are allowed to complete their work before the server returns control to the client which requested the global read lock. The delayed insert handler will close the table and finish when the outstanding inserts are done. */ if (! (thd->lock= mysql_lock_tables(thd, &di->table, 1, 0))) { /* Fatal error */ thd->killed= THD::KILL_CONNECTION; } mysql_cond_broadcast(&di->cond_client); } if (di->stacked_inserts) { if (di->handle_inserts()) { /* Some fatal error */ thd->killed= THD::KILL_CONNECTION; } } di->status=0; if (!di->stacked_inserts && !di->tables_in_use && thd->lock) { /*
No one is doing a insert delayed Unlock table so that other threads can use it */ MYSQL_LOCK *lock=thd->lock; thd->lock=0; mysql_mutex_unlock(&di->mutex); /*
We need to release next_insert_id before unlocking. This is enforced by handler::ha_external_lock(). */ di->table->file->ha_release_auto_increment(); mysql_unlock_tables(thd, lock); trans_commit_stmt(thd); di->group_count=0; mysql_audit_release(thd); mysql_mutex_lock(&di->mutex); } if (di->tables_in_use) mysql_cond_broadcast(&di->cond_client); // If waiting clients
}
err: DBUG_LEAVE; }
close_thread_tables(thd); // Free the table
thd->mdl_context.release_transactional_locks(); di->table=0; thd->killed= THD::KILL_CONNECTION; // If error
mysql_cond_broadcast(&di->cond_client); // Safety
mysql_mutex_unlock(&di->mutex);
mysql_mutex_lock(&LOCK_delayed_create); // Because of delayed_get_table
mysql_mutex_lock(&LOCK_delayed_insert); /*
di should be unlinked from the thread handler list and have no active clients */ delete di; mysql_mutex_unlock(&LOCK_delayed_insert); mysql_mutex_unlock(&LOCK_delayed_create);
my_thread_end(); pthread_exit(0);
return 0;}
/* Remove pointers from temporary fields to allocated values */
static void unlink_blobs(register TABLE *table){ for (Field **ptr=table->field ; *ptr ; ptr++) { if ((*ptr)->flags & BLOB_FLAG) ((Field_blob *) (*ptr))->clear_temporary(); }}
/* Free blobs stored in current row */
static void free_delayed_insert_blobs(register TABLE *table){ for (Field **ptr=table->field ; *ptr ; ptr++) { if ((*ptr)->flags & BLOB_FLAG) { uchar *str; ((Field_blob *) (*ptr))->get_ptr(&str); my_free(str); ((Field_blob *) (*ptr))->reset(); } }}
bool Delayed_insert::handle_inserts(void){ int error; ulong max_rows; bool has_trans = TRUE; bool using_ignore= 0, using_opt_replace= 0, using_bin_log= mysql_bin_log.is_open(); delayed_row *row; DBUG_ENTER("handle_inserts");
/* Allow client to insert new rows */ mysql_mutex_unlock(&mutex);
table->next_number_field=table->found_next_number_field; table->use_all_columns();
thd_proc_info(&thd, "upgrading lock"); if (thr_upgrade_write_delay_lock(*thd.lock->locks, delayed_lock, thd.variables.lock_wait_timeout)) { /*
This can happen if thread is killed either by a shutdown or if another thread is removing the current table definition from the table cache. */ my_error(ER_DELAYED_CANT_CHANGE_LOCK,MYF(ME_FATALERROR), table->s->table_name.str); goto err; }
thd_proc_info(&thd, "insert"); max_rows= delayed_insert_limit; if (thd.killed || table->s->has_old_version()) { thd.killed= THD::KILL_CONNECTION; max_rows= ULONG_MAX; // Do as much as possible
}
/*
We can't use row caching when using the binary log because if we get a crash, then binary log will contain rows that are not yet written to disk, which will cause problems in replication. */ if (!using_bin_log) table->file->extra(HA_EXTRA_WRITE_CACHE); mysql_mutex_lock(&mutex);
while ((row=rows.get())) { stacked_inserts--; mysql_mutex_unlock(&mutex); memcpy(table->record[0],row->record,table->s->reclength);
thd.start_time=row->start_time; thd.query_start_used=row->query_start_used; /*
To get the exact auto_inc interval to store in the binlog we must not use values from the previous interval (of the previous rows). */ bool log_query= (row->log_query && row->query.str != NULL); DBUG_PRINT("delayed", ("query: '%s' length: %lu", row->query.str ? row->query.str : "[NULL]", (ulong) row->query.length)); if (log_query) { /*
This is the first value of an INSERT statement. It is the right place to clear a forced insert_id. This is usually done after the last value of an INSERT statement, but we won't know this in the insert delayed thread. But before the first value is sufficiently equivalent to after the last value of the previous statement. */ table->file->ha_release_auto_increment(); thd.auto_inc_intervals_in_cur_stmt_for_binlog.empty(); } thd.first_successful_insert_id_in_prev_stmt= row->first_successful_insert_id_in_prev_stmt; thd.stmt_depends_on_first_successful_insert_id_in_prev_stmt= row->stmt_depends_on_first_successful_insert_id_in_prev_stmt; table->timestamp_field_type= row->timestamp_field_type; table->auto_increment_field_not_null= row->auto_increment_field_not_null;
/* Copy the session variables. */ thd.variables.auto_increment_increment= row->auto_increment_increment; thd.variables.auto_increment_offset= row->auto_increment_offset; thd.variables.sql_mode= row->sql_mode;
/* Copy a forced insert_id, if any. */ if (row->forced_insert_id) { DBUG_PRINT("delayed", ("received auto_inc: %lu", (ulong) row->forced_insert_id)); thd.force_one_auto_inc_interval(row->forced_insert_id); }
info.ignore= row->ignore; info.handle_duplicates= row->dup; if (info.ignore || info.handle_duplicates != DUP_ERROR) { table->file->extra(HA_EXTRA_IGNORE_DUP_KEY); using_ignore=1; } if (info.handle_duplicates == DUP_REPLACE && (!table->triggers || !table->triggers->has_delete_triggers())) { table->file->extra(HA_EXTRA_WRITE_CAN_REPLACE); using_opt_replace= 1; } if (info.handle_duplicates == DUP_UPDATE) table->file->extra(HA_EXTRA_INSERT_WITH_UPDATE); thd.clear_error(); // reset error for binlog
if (write_record(&thd, table, &info)) { info.error_count++; // Ignore errors
thread_safe_increment(delayed_insert_errors,&LOCK_delayed_status); row->log_query = 0; }
if (using_ignore) { using_ignore=0; table->file->extra(HA_EXTRA_NO_IGNORE_DUP_KEY); } if (using_opt_replace) { using_opt_replace= 0; table->file->extra(HA_EXTRA_WRITE_CANNOT_REPLACE); }
if (log_query && mysql_bin_log.is_open()) { bool backup_time_zone_used = thd.time_zone_used; Time_zone *backup_time_zone = thd.variables.time_zone; if (row->time_zone != NULL) { thd.time_zone_used = true; thd.variables.time_zone = row->time_zone; }
/* if the delayed insert was killed, the killed status is
ignored while binlogging */ int errcode= 0; if (thd.killed == THD::NOT_KILLED) errcode= query_error_code(&thd, TRUE); /*
If the query has several rows to insert, only the first row will come here. In row-based binlogging, this means that the first row will be written to binlog as one Table_map event and one Rows event (due to an event flush done in binlog_query()), then all other rows of this query will be binlogged together as one single Table_map event and one single Rows event. */ if (thd.binlog_query(THD::ROW_QUERY_TYPE, row->query.str, row->query.length, FALSE, FALSE, FALSE, errcode)) goto err;
thd.time_zone_used = backup_time_zone_used; thd.variables.time_zone = backup_time_zone; }
if (table->s->blob_fields) free_delayed_insert_blobs(table); thread_safe_decrement(delayed_rows_in_use,&LOCK_delayed_status); thread_safe_increment(delayed_insert_writes,&LOCK_delayed_status); mysql_mutex_lock(&mutex);
/*
Reset the table->auto_increment_field_not_null as it is valid for only one row. */ table->auto_increment_field_not_null= FALSE;
delete row; /*
Let READ clients do something once in a while We should however not break in the middle of a multi-line insert if we have binary logging enabled as we don't want other commands on this table until all entries has been processed */ if (group_count++ >= max_rows && (row= rows.head()) && (!(row->log_query & using_bin_log))) { group_count=0; if (stacked_inserts || tables_in_use) // Let these wait a while
{ if (tables_in_use) mysql_cond_broadcast(&cond_client); // If waiting clients
thd_proc_info(&thd, "reschedule"); mysql_mutex_unlock(&mutex); if ((error=table->file->extra(HA_EXTRA_NO_CACHE))) { /* This should never happen */ table->file->print_error(error,MYF(0)); sql_print_error("%s", thd.stmt_da->message()); DBUG_PRINT("error", ("HA_EXTRA_NO_CACHE failed in loop")); goto err; } query_cache_invalidate3(&thd, table, 1); if (thr_reschedule_write_lock(*thd.lock->locks, thd.variables.lock_wait_timeout)) { /* This is not known to happen. */ my_error(ER_DELAYED_CANT_CHANGE_LOCK,MYF(ME_FATALERROR), table->s->table_name.str); goto err; } if (!using_bin_log) table->file->extra(HA_EXTRA_WRITE_CACHE); mysql_mutex_lock(&mutex); thd_proc_info(&thd, "insert"); } if (tables_in_use) mysql_cond_broadcast(&cond_client); // If waiting clients
} } thd_proc_info(&thd, 0); mysql_mutex_unlock(&mutex);
/*
We need to flush the pending event when using row-based replication since the flushing normally done in binlog_query() is not done last in the statement: for delayed inserts, the insert statement is logged *before* all rows are inserted.
We can flush the pending event without checking the thd->lock since the delayed insert *thread* is not inside a stored function or trigger.
TODO: Move the logging to last in the sequence of rows. */ has_trans= thd.lex->sql_command == SQLCOM_CREATE_TABLE || table->file->has_transactions(); if (thd.is_current_stmt_binlog_format_row() && thd.binlog_flush_pending_rows_event(TRUE, has_trans)) goto err;
if ((error=table->file->extra(HA_EXTRA_NO_CACHE))) { // This shouldn't happen
table->file->print_error(error,MYF(0)); sql_print_error("%s", thd.stmt_da->message()); DBUG_PRINT("error", ("HA_EXTRA_NO_CACHE failed after loop")); goto err; } query_cache_invalidate3(&thd, table, 1); mysql_mutex_lock(&mutex); DBUG_RETURN(0);
err:#ifndef DBUG_OFF
max_rows= 0; // For DBUG output
#endif
/* Remove all not used rows */ while ((row=rows.get())) { if (table->s->blob_fields) { memcpy(table->record[0],row->record,table->s->reclength); free_delayed_insert_blobs(table); } delete row; thread_safe_increment(delayed_insert_errors,&LOCK_delayed_status); stacked_inserts--;#ifndef DBUG_OFF
max_rows++;#endif
} DBUG_PRINT("error", ("dropped %lu rows after an error", max_rows)); thread_safe_increment(delayed_insert_errors, &LOCK_delayed_status); mysql_mutex_lock(&mutex); DBUG_RETURN(1);}#endif /* EMBEDDED_LIBRARY */
/***************************************************************************
Store records in INSERT ... SELECT ****************************************************************************/
/*
make insert specific preparation and checks after opening tables
SYNOPSIS mysql_insert_select_prepare() thd thread handler
RETURN FALSE OK TRUE Error*/
bool mysql_insert_select_prepare(THD *thd){ LEX *lex= thd->lex; SELECT_LEX *select_lex= &lex->select_lex; TABLE_LIST *first_select_leaf_table; DBUG_ENTER("mysql_insert_select_prepare");
/*
SELECT_LEX do not belong to INSERT statement, so we can't add WHERE clause if table is VIEW */ if (mysql_prepare_insert(thd, lex->query_tables, lex->query_tables->table, lex->field_list, 0, lex->update_list, lex->value_list, lex->duplicates, &select_lex->where, TRUE, FALSE, FALSE)) DBUG_RETURN(TRUE);
/*
exclude first table from leaf tables list, because it belong to INSERT */ DBUG_ASSERT(select_lex->leaf_tables != 0); lex->leaf_tables_insert= select_lex->leaf_tables; /* skip all leaf tables belonged to view where we are insert */ for (first_select_leaf_table= select_lex->leaf_tables->next_leaf; first_select_leaf_table && first_select_leaf_table->belong_to_view && first_select_leaf_table->belong_to_view == lex->leaf_tables_insert->belong_to_view; first_select_leaf_table= first_select_leaf_table->next_leaf) {} select_lex->leaf_tables= first_select_leaf_table; DBUG_RETURN(FALSE);}
select_insert::select_insert(TABLE_LIST *table_list_par, TABLE *table_par, List<Item> *fields_par, List<Item> *update_fields, List<Item> *update_values, enum_duplicates duplic, bool ignore_check_option_errors) :table_list(table_list_par), table(table_par), fields(fields_par), autoinc_value_of_last_inserted_row(0), insert_into_view(table_list_par && table_list_par->view != 0){ bzero((char*) &info,sizeof(info)); info.handle_duplicates= duplic; info.ignore= ignore_check_option_errors; info.update_fields= update_fields; info.update_values= update_values; if (table_list_par) info.view= (table_list_par->view ? table_list_par : 0);}
intselect_insert::prepare(List<Item> &values, SELECT_LEX_UNIT *u){ LEX *lex= thd->lex; int res; table_map map= 0; SELECT_LEX *lex_current_select_save= lex->current_select; DBUG_ENTER("select_insert::prepare");
unit= u;
/*
Since table in which we are going to insert is added to the first select, LEX::current_select should point to the first select while we are fixing fields from insert list. */ lex->current_select= &lex->select_lex;
/* Errors during check_insert_fields() should not be ignored. */ lex->current_select->no_error= FALSE; res= (setup_fields(thd, 0, values, MARK_COLUMNS_READ, 0, 0) || check_insert_fields(thd, table_list, *fields, values, !insert_into_view, 1, &map));
if (!res && fields->elements) { bool saved_abort_on_warning= thd->abort_on_warning; thd->abort_on_warning= !info.ignore && (thd->variables.sql_mode & (MODE_STRICT_TRANS_TABLES | MODE_STRICT_ALL_TABLES)); res= check_that_all_fields_are_given_values(thd, table_list->table, table_list); thd->abort_on_warning= saved_abort_on_warning; }
if (info.handle_duplicates == DUP_UPDATE && !res) { Name_resolution_context *context= &lex->select_lex.context; Name_resolution_context_state ctx_state;
/* Save the state of the current name resolution context. */ ctx_state.save_state(context, table_list);
/* Perform name resolution only in the first table - 'table_list'. */ table_list->next_local= 0; context->resolve_in_table_list_only(table_list);
lex->select_lex.no_wrap_view_item= TRUE; res= res || check_update_fields(thd, context->table_list, *info.update_fields, *info.update_values, &map); lex->select_lex.no_wrap_view_item= FALSE; /*
When we are not using GROUP BY and there are no ungrouped aggregate functions we can refer to other tables in the ON DUPLICATE KEY part. We use next_name_resolution_table descructively, so check it first (views?) */ DBUG_ASSERT (!table_list->next_name_resolution_table); if (lex->select_lex.group_list.elements == 0 && !lex->select_lex.with_sum_func) /*
We must make a single context out of the two separate name resolution contexts : the INSERT table and the tables in the SELECT part of INSERT ... SELECT. To do that we must concatenate the two lists */ table_list->next_name_resolution_table= ctx_state.get_first_name_resolution_table();
res= res || setup_fields(thd, 0, *info.update_values, MARK_COLUMNS_READ, 0, 0); if (!res) { /*
Traverse the update values list and substitute fields from the select for references (Item_ref objects) to them. This is done in order to get correct values from those fields when the select employs a temporary table. */ List_iterator<Item> li(*info.update_values); Item *item;
while ((item= li++)) { item->transform(&Item::update_value_transformer, (uchar*)lex->current_select); } }
/* Restore the current context. */ ctx_state.restore_state(context, table_list); }
lex->current_select= lex_current_select_save; if (res) DBUG_RETURN(1); /*
if it is INSERT into join view then check_insert_fields already found real table for insert */ table= table_list->table;
/*
Is table which we are changing used somewhere in other parts of query */ if (unique_table(thd, table_list, table_list->next_global, 0)) { /* Using same table for INSERT and SELECT */ lex->current_select->options|= OPTION_BUFFER_RESULT; lex->current_select->join->select_options|= OPTION_BUFFER_RESULT; } else if (!(lex->current_select->options & OPTION_BUFFER_RESULT) && thd->locked_tables_mode <= LTM_LOCK_TABLES) { /*
We must not yet prepare the result table if it is the same as one of the source tables (INSERT SELECT). The preparation may disable indexes on the result table, which may be used during the select, if it is the same table (Bug #6034). Do the preparation after the select phase in select_insert::prepare2(). We won't start bulk inserts at all if this statement uses functions or should invoke triggers since they may access to the same table too. */ table->file->ha_start_bulk_insert((ha_rows) 0); } restore_record(table,s->default_values); // Get empty record
table->next_number_field=table->found_next_number_field;
#ifdef HAVE_REPLICATION
if (thd->slave_thread && (info.handle_duplicates == DUP_UPDATE) && (table->next_number_field != NULL) && rpl_master_has_bug(&active_mi->rli, 24432, TRUE, NULL, NULL)) DBUG_RETURN(1);#endif
thd->cuted_fields=0; if (info.ignore || info.handle_duplicates != DUP_ERROR) table->file->extra(HA_EXTRA_IGNORE_DUP_KEY); if (info.handle_duplicates == DUP_REPLACE && (!table->triggers || !table->triggers->has_delete_triggers())) table->file->extra(HA_EXTRA_WRITE_CAN_REPLACE); if (info.handle_duplicates == DUP_UPDATE) table->file->extra(HA_EXTRA_INSERT_WITH_UPDATE); thd->abort_on_warning= (!info.ignore && (thd->variables.sql_mode & (MODE_STRICT_TRANS_TABLES | MODE_STRICT_ALL_TABLES))); res= (table_list->prepare_where(thd, 0, TRUE) || table_list->prepare_check_option(thd));
if (!res) prepare_triggers_for_insert_stmt(table);
DBUG_RETURN(res);}
/*
Finish the preparation of the result table.
SYNOPSIS select_insert::prepare2() void
DESCRIPTION If the result table is the same as one of the source tables (INSERT SELECT), the result table is not finally prepared at the join prepair phase. Do the final preparation now. RETURN 0 OK*/
int select_insert::prepare2(void){ DBUG_ENTER("select_insert::prepare2"); if (thd->lex->current_select->options & OPTION_BUFFER_RESULT && thd->locked_tables_mode <= LTM_LOCK_TABLES) table->file->ha_start_bulk_insert((ha_rows) 0); DBUG_RETURN(0);}
void select_insert::cleanup(){ /* select_insert/select_create are never re-used in prepared statement */ DBUG_ASSERT(0);}
select_insert::~select_insert(){ DBUG_ENTER("~select_insert"); if (table) { table->next_number_field=0; table->auto_increment_field_not_null= FALSE; table->file->ha_reset(); } thd->count_cuted_fields= CHECK_FIELD_IGNORE; thd->abort_on_warning= 0; DBUG_VOID_RETURN;}
bool select_insert::send_data(List<Item> &values){ DBUG_ENTER("select_insert::send_data"); bool error=0;
if (unit->offset_limit_cnt) { // using limit offset,count
unit->offset_limit_cnt--; DBUG_RETURN(0); }
thd->count_cuted_fields= CHECK_FIELD_WARN; // Calculate cuted fields
store_values(values); thd->count_cuted_fields= CHECK_FIELD_ERROR_FOR_NULL; if (thd->is_error()) { table->auto_increment_field_not_null= FALSE; DBUG_RETURN(1); } if (table_list) // Not CREATE ... SELECT
{ switch (table_list->view_check_option(thd, info.ignore)) { case VIEW_CHECK_SKIP: DBUG_RETURN(0); case VIEW_CHECK_ERROR: DBUG_RETURN(1); } }
// Release latches in case bulk insert takes a long time
ha_release_temporary_latches(thd);
error= write_record(thd, table, &info); table->auto_increment_field_not_null= FALSE; if (!error) { if (table->triggers || info.handle_duplicates == DUP_UPDATE) { /*
Restore fields of the record since it is possible that they were changed by ON DUPLICATE KEY UPDATE clause. If triggers exist then whey can modify some fields which were not originally touched by INSERT ... SELECT, so we have to restore their original values for the next row. */ restore_record(table, s->default_values); } if (table->next_number_field) { /*
If no value has been autogenerated so far, we need to remember the value we just saw, we may need to send it to client in the end. */ if (thd->first_successful_insert_id_in_cur_stmt == 0) // optimization
autoinc_value_of_last_inserted_row= table->next_number_field->val_int(); /*
Clear auto-increment field for the next record, if triggers are used we will clear it twice, but this should be cheap. */ table->next_number_field->reset(); } } DBUG_RETURN(error);}
void select_insert::store_values(List<Item> &values){ if (fields->elements) fill_record_n_invoke_before_triggers(thd, *fields, values, 1, table->triggers, TRG_EVENT_INSERT); else fill_record_n_invoke_before_triggers(thd, table->field, values, 1, table->triggers, TRG_EVENT_INSERT);}
void select_insert::send_error(uint errcode,const char *err){ DBUG_ENTER("select_insert::send_error");
my_message(errcode, err, MYF(0));
DBUG_VOID_RETURN;}
bool select_insert::send_eof(){ int error; bool const trans_table= table->file->has_transactions(); ulonglong id, row_count; bool changed; THD::killed_state killed_status= thd->killed; DBUG_ENTER("select_insert::send_eof"); DBUG_PRINT("enter", ("trans_table=%d, table_type='%s'", trans_table, table->file->table_type()));
error= (thd->locked_tables_mode <= LTM_LOCK_TABLES ? table->file->ha_end_bulk_insert() : 0); table->file->extra(HA_EXTRA_NO_IGNORE_DUP_KEY); table->file->extra(HA_EXTRA_WRITE_CANNOT_REPLACE);
changed= (info.copied || info.deleted || info.updated); if (changed) { /*
We must invalidate the table in the query cache before binlog writing and ha_autocommit_or_rollback. */ query_cache_invalidate3(thd, table, 1); }
if (thd->transaction.stmt.modified_non_trans_table) thd->transaction.all.modified_non_trans_table= TRUE;
DBUG_ASSERT(trans_table || !changed || thd->transaction.stmt.modified_non_trans_table);
/*
Write to binlog before commiting transaction. No statement will be written by the binlog_query() below in RBR mode. All the events are in the transaction cache and will be written when ha_autocommit_or_rollback() is issued below. */ if (mysql_bin_log.is_open() && (!error || thd->transaction.stmt.modified_non_trans_table)) { int errcode= 0; if (!error) thd->clear_error(); else errcode= query_error_code(thd, killed_status == THD::NOT_KILLED); if (thd->binlog_query(THD::ROW_QUERY_TYPE, thd->query(), thd->query_length(), trans_table, FALSE, FALSE, errcode)) { table->file->ha_release_auto_increment(); DBUG_RETURN(1); } } table->file->ha_release_auto_increment();
if (error) { table->file->print_error(error,MYF(0)); DBUG_RETURN(1); } char buff[160]; if (info.ignore) sprintf(buff, ER(ER_INSERT_INFO), (ulong) info.records, (ulong) (info.records - info.copied), (ulong) thd->warning_info->statement_warn_count()); else sprintf(buff, ER(ER_INSERT_INFO), (ulong) info.records, (ulong) (info.deleted+info.updated), (ulong) thd->warning_info->statement_warn_count()); row_count= info.copied + info.deleted + ((thd->client_capabilities & CLIENT_FOUND_ROWS) ? info.touched : info.updated); id= (thd->first_successful_insert_id_in_cur_stmt > 0) ? thd->first_successful_insert_id_in_cur_stmt : (thd->arg_of_last_insert_id_function ? thd->first_successful_insert_id_in_prev_stmt : (info.copied ? autoinc_value_of_last_inserted_row : 0)); ::my_ok(thd, row_count, id, buff); DBUG_RETURN(0);}
void select_insert::abort_result_set() {
DBUG_ENTER("select_insert::abort_result_set"); /*
If the creation of the table failed (due to a syntax error, for example), no table will have been opened and therefore 'table' will be NULL. In that case, we still need to execute the rollback and the end of the function. */ if (table) { bool changed, transactional_table; /*
If we are not in prelocked mode, we end the bulk insert started before. */ if (thd->locked_tables_mode <= LTM_LOCK_TABLES) table->file->ha_end_bulk_insert();
/*
If at least one row has been inserted/modified and will stay in the table (the table doesn't have transactions) we must write to the binlog (and the error code will make the slave stop).
For many errors (example: we got a duplicate key error while inserting into a MyISAM table), no row will be added to the table, so passing the error to the slave will not help since there will be an error code mismatch (the inserts will succeed on the slave with no error).
If table creation failed, the number of rows modified will also be zero, so no check for that is made. */ changed= (info.copied || info.deleted || info.updated); transactional_table= table->file->has_transactions(); if (thd->transaction.stmt.modified_non_trans_table) { if (!can_rollback_data()) thd->transaction.all.modified_non_trans_table= TRUE;
if (mysql_bin_log.is_open()) { int errcode= query_error_code(thd, thd->killed == THD::NOT_KILLED); /* error of writing binary log is ignored */ (void) thd->binlog_query(THD::ROW_QUERY_TYPE, thd->query(), thd->query_length(), transactional_table, FALSE, FALSE, errcode); } if (changed) query_cache_invalidate3(thd, table, 1); } DBUG_ASSERT(transactional_table || !changed || thd->transaction.stmt.modified_non_trans_table); table->file->ha_release_auto_increment(); }
DBUG_VOID_RETURN;}
/***************************************************************************
CREATE TABLE (SELECT) ...***************************************************************************/
/**
Create table from lists of fields and items (or just return TABLE object for pre-opened existing table).
@param thd [in] Thread object @param create_info [in] Create information (like MAX_ROWS, ENGINE or temporary table flag) @param create_table [in] Pointer to TABLE_LIST object providing database and name for table to be created or to be open @param alter_info [in/out] Initial list of columns and indexes for the table to be created @param items [in] List of items which should be used to produce rest of fields for the table (corresponding fields will be added to the end of alter_info->create_list) @param lock [out] Pointer to the MYSQL_LOCK object for table created will be returned in this parameter. Since this table is not included in THD::lock caller is responsible for explicitly unlocking this table. @param hooks [in] Hooks to be invoked before and after obtaining table lock on the table being created.
@note This function assumes that either table exists and was pre-opened and locked at open_and_lock_tables() stage (and in this case we just emit error or warning and return pre-opened TABLE object) or an exclusive metadata lock was acquired on table so we can safely create, open and lock table in it (we don't acquire metadata lock if this create is for temporary table).
@note Since this function contains some logic specific to CREATE TABLE ... SELECT it should be changed before it can be used in other contexts.
@retval non-zero Pointer to TABLE object for table created or opened @retval 0 Error*/
static TABLE *create_table_from_items(THD *thd, HA_CREATE_INFO *create_info, TABLE_LIST *create_table, Alter_info *alter_info, List<Item> *items, MYSQL_LOCK **lock, TABLEOP_HOOKS *hooks){ TABLE tmp_table; // Used during 'Create_field()'
TABLE_SHARE share; TABLE *table= 0; uint select_field_count= items->elements; /* Add selected items to field list */ List_iterator_fast<Item> it(*items); Item *item; Field *tmp_field; DBUG_ENTER("create_table_from_items");
tmp_table.alias= 0; tmp_table.timestamp_field= 0; tmp_table.s= &share; init_tmp_table_share(thd, &share, "", 0, "", "");
tmp_table.s->db_create_options=0; tmp_table.s->blob_ptr_size= portable_sizeof_char_ptr; tmp_table.s->db_low_byte_first= test(create_info->db_type == myisam_hton || create_info->db_type == heap_hton); tmp_table.null_row=tmp_table.maybe_null=0;
while ((item=it++)) { Create_field *cr_field; Field *field, *def_field; if (item->type() == Item::FUNC_ITEM) if (item->result_type() != STRING_RESULT) field= item->tmp_table_field(&tmp_table); else field= item->tmp_table_field_from_field_type(&tmp_table, 0); else field= create_tmp_field(thd, &tmp_table, item, item->type(), (Item ***) 0, &tmp_field, &def_field, 0, 0, 0, 0, 0); if (!field || !(cr_field=new Create_field(field,(item->type() == Item::FIELD_ITEM ? ((Item_field *)item)->field : (Field*) 0)))) DBUG_RETURN(0); if (item->maybe_null) cr_field->flags &= ~NOT_NULL_FLAG; alter_info->create_list.push_back(cr_field); }
DBUG_EXECUTE_IF("sleep_create_select_before_create", my_sleep(6000000););
/*
Create and lock table.
Note that we either creating (or opening existing) temporary table or creating base table on which name we have exclusive lock. So code below should not cause deadlocks or races.
We don't log the statement, it will be logged later.
If this is a HEAP table, the automatic DELETE FROM which is written to the binlog when a HEAP table is opened for the first time since startup, must not be written: 1) it would be wrong (imagine we're in CREATE SELECT: we don't want to delete from it) 2) it would be written before the CREATE TABLE, which is a wrong order. So we keep binary logging disabled when we open_table(). */ { if (!mysql_create_table_no_lock(thd, create_table->db, create_table->table_name, create_info, alter_info, 0, select_field_count, NULL)) { DBUG_EXECUTE_IF("sleep_create_select_before_open", my_sleep(6000000););
if (!(create_info->options & HA_LEX_CREATE_TMP_TABLE)) { Open_table_context ot_ctx(thd, MYSQL_OPEN_REOPEN); /*
Here we open the destination table, on which we already have an exclusive metadata lock. */ if (open_table(thd, create_table, thd->mem_root, &ot_ctx)) { quick_rm_table(create_info->db_type, create_table->db, table_case_name(create_info, create_table->table_name), 0); } else table= create_table->table; } else { Open_table_context ot_ctx(thd, MYSQL_OPEN_TEMPORARY_ONLY); if (open_table(thd, create_table, thd->mem_root, &ot_ctx)) { /*
This shouldn't happen as creation of temporary table should make it preparable for open. But let us do close_temporary_table() here just in case. */ drop_temporary_table(thd, create_table, NULL); } else table= create_table->table; } } if (!table) // open failed
DBUG_RETURN(0); }
DBUG_EXECUTE_IF("sleep_create_select_before_lock", my_sleep(6000000););
table->reginfo.lock_type=TL_WRITE; hooks->prelock(&table, 1); // Call prelock hooks
/*
mysql_lock_tables() below should never fail with request to reopen table since it won't wait for the table lock (we have exclusive metadata lock on the table) and thus can't get aborted. */ if (! ((*lock)= mysql_lock_tables(thd, &table, 1, 0)) || hooks->postlock(&table, 1)) { if (*lock) { mysql_unlock_tables(thd, *lock); *lock= 0; } drop_open_table(thd, table, create_table->db, create_table->table_name); DBUG_RETURN(0); } DBUG_RETURN(table);}
intselect_create::prepare(List<Item> &values, SELECT_LEX_UNIT *u){ MYSQL_LOCK *extra_lock= NULL; DBUG_ENTER("select_create::prepare");
TABLEOP_HOOKS *hook_ptr= NULL; /*
For row-based replication, the CREATE-SELECT statement is written in two pieces: the first one contain the CREATE TABLE statement necessary to create the table and the second part contain the rows that should go into the table.
For non-temporary tables, the start of the CREATE-SELECT implicitly commits the previous transaction, and all events forming the statement will be stored the transaction cache. At end of the statement, the entire statement is committed as a transaction, and all events are written to the binary log.
On the master, the table is locked for the duration of the statement, but since the CREATE part is replicated as a simple statement, there is no way to lock the table for accesses on the slave. Hence, we have to hold on to the CREATE part of the statement until the statement has finished. */ class MY_HOOKS : public TABLEOP_HOOKS { public: MY_HOOKS(select_create *x, TABLE_LIST *create_table_arg, TABLE_LIST *select_tables_arg) : ptr(x), create_table(create_table_arg), select_tables(select_tables_arg) { }
private: virtual int do_postlock(TABLE **tables, uint count) { int error; THD *thd= const_cast<THD*>(ptr->get_thd()); TABLE_LIST *save_next_global= create_table->next_global;
create_table->next_global= select_tables;
error= thd->decide_logging_format(create_table);
create_table->next_global= save_next_global;
if (error) return error;
TABLE const *const table = *tables; if (thd->is_current_stmt_binlog_format_row() && !table->s->tmp_table) { if (int error= ptr->binlog_show_create_table(tables, count)) return error; } return 0; } select_create *ptr; TABLE_LIST *create_table; TABLE_LIST *select_tables; };
MY_HOOKS hooks(this, create_table, select_tables); hook_ptr= &hooks;
unit= u;
/*
Start a statement transaction before the create if we are using row-based replication for the statement. If we are creating a temporary table, we need to start a statement transaction. */ if ((thd->lex->create_info.options & HA_LEX_CREATE_TMP_TABLE) == 0 && thd->is_current_stmt_binlog_format_row() && mysql_bin_log.is_open()) { thd->binlog_start_trans_and_stmt(); }
DBUG_ASSERT(create_table->table == NULL);
DBUG_EXECUTE_IF("sleep_create_select_before_check_if_exists", my_sleep(6000000););
if (!(table= create_table_from_items(thd, create_info, create_table, alter_info, &values, &extra_lock, hook_ptr))) /* abort() deletes table */ DBUG_RETURN(-1);
if (extra_lock) { DBUG_ASSERT(m_plock == NULL);
if (create_info->options & HA_LEX_CREATE_TMP_TABLE) m_plock= &m_lock; else m_plock= &thd->extra_lock;
*m_plock= extra_lock; }
if (table->s->fields < values.elements) { my_error(ER_WRONG_VALUE_COUNT_ON_ROW, MYF(0), 1); DBUG_RETURN(-1); }
/* First field to copy */ field= table->field+table->s->fields - values.elements;
/* Mark all fields that are given values */ for (Field **f= field ; *f ; f++) bitmap_set_bit(table->write_set, (*f)->field_index);
/* Don't set timestamp if used */ table->timestamp_field_type= TIMESTAMP_NO_AUTO_SET; table->next_number_field=table->found_next_number_field;
restore_record(table,s->default_values); // Get empty record
thd->cuted_fields=0; if (info.ignore || info.handle_duplicates != DUP_ERROR) table->file->extra(HA_EXTRA_IGNORE_DUP_KEY); if (info.handle_duplicates == DUP_REPLACE && (!table->triggers || !table->triggers->has_delete_triggers())) table->file->extra(HA_EXTRA_WRITE_CAN_REPLACE); if (info.handle_duplicates == DUP_UPDATE) table->file->extra(HA_EXTRA_INSERT_WITH_UPDATE); if (thd->locked_tables_mode <= LTM_LOCK_TABLES) table->file->ha_start_bulk_insert((ha_rows) 0); thd->abort_on_warning= (!info.ignore && (thd->variables.sql_mode & (MODE_STRICT_TRANS_TABLES | MODE_STRICT_ALL_TABLES))); if (check_that_all_fields_are_given_values(thd, table, table_list)) DBUG_RETURN(1); table->mark_columns_needed_for_insert(); table->file->extra(HA_EXTRA_WRITE_CACHE); DBUG_RETURN(0);}
intselect_create::binlog_show_create_table(TABLE **tables, uint count){ /*
Note 1: In RBR mode, we generate a CREATE TABLE statement for the created table by calling store_create_info() (behaves as SHOW CREATE TABLE). In the event of an error, nothing should be written to the binary log, even if the table is non-transactional; therefore we pretend that the generated CREATE TABLE statement is for a transactional table. The event will then be put in the transaction cache, and any subsequent events (e.g., table-map events and binrow events) will also be put there. We can then use ha_autocommit_or_rollback() to either throw away the entire kaboodle of events, or write them to the binary log.
We write the CREATE TABLE statement here and not in prepare() since there potentially are sub-selects or accesses to information schema that will do a close_thread_tables(), destroying the statement transaction cache. */ DBUG_ASSERT(thd->is_current_stmt_binlog_format_row()); DBUG_ASSERT(tables && *tables && count > 0);
char buf[2048]; String query(buf, sizeof(buf), system_charset_info); int result; TABLE_LIST tmp_table_list;
memset(&tmp_table_list, 0, sizeof(tmp_table_list)); tmp_table_list.table = *tables; query.length(0); // Have to zero it since constructor doesn't
result= store_create_info(thd, &tmp_table_list, &query, create_info, /* show_database */ TRUE); DBUG_ASSERT(result == 0); /* store_create_info() always return 0 */
if (mysql_bin_log.is_open()) { int errcode= query_error_code(thd, thd->killed == THD::NOT_KILLED); result= thd->binlog_query(THD::STMT_QUERY_TYPE, query.ptr(), query.length(), /* is_trans */ TRUE, /* direct */ FALSE, /* suppress_use */ FALSE, errcode); } return result;}
void select_create::store_values(List<Item> &values){ fill_record_n_invoke_before_triggers(thd, field, values, 1, table->triggers, TRG_EVENT_INSERT);}
void select_create::send_error(uint errcode,const char *err){ DBUG_ENTER("select_create::send_error");
DBUG_PRINT("info", ("Current statement %s row-based", thd->is_current_stmt_binlog_format_row() ? "is" : "is NOT")); DBUG_PRINT("info", ("Current table (at 0x%lu) %s a temporary (or non-existant) table", (ulong) table, table && !table->s->tmp_table ? "is NOT" : "is")); /*
This will execute any rollbacks that are necessary before writing the transcation cache.
We disable the binary log since nothing should be written to the binary log. This disabling is important, since we potentially do a "roll back" of non-transactional tables by removing the table, and the actual rollback might generate events that should not be written to the binary log.
*/ tmp_disable_binlog(thd); select_insert::send_error(errcode, err); reenable_binlog(thd);
DBUG_VOID_RETURN;}
bool select_create::send_eof(){ bool tmp=select_insert::send_eof(); if (tmp) abort(); else { /*
Do an implicit commit at end of statement for non-temporary tables. This can fail, but we should unlock the table nevertheless. */ if (!table->s->tmp_table) { trans_commit_stmt(thd); trans_commit_implicit(thd); }
table->file->extra(HA_EXTRA_NO_IGNORE_DUP_KEY); table->file->extra(HA_EXTRA_WRITE_CANNOT_REPLACE); if (m_plock) { mysql_unlock_tables(thd, *m_plock); *m_plock= NULL; m_plock= NULL; } } return tmp;}
void select_create::abort_result_set(){ DBUG_ENTER("select_create::abort_result_set");
/*
In select_insert::abort() we roll back the statement, including truncating the transaction cache of the binary log. To do this, we pretend that the statement is transactional, even though it might be the case that it was not.
We roll back the statement prior to deleting the table and prior to releasing the lock on the table, since there might be potential for failure if the rollback is executed after the drop or after unlocking the table.
We also roll back the statement regardless of whether the creation of the table succeeded or not, since we need to reset the binary log state. */ tmp_disable_binlog(thd); select_insert::abort_result_set(); thd->transaction.stmt.modified_non_trans_table= FALSE; reenable_binlog(thd); /* possible error of writing binary log is ignored deliberately */ (void) thd->binlog_flush_pending_rows_event(TRUE, TRUE);
if (m_plock) { mysql_unlock_tables(thd, *m_plock); *m_plock= NULL; m_plock= NULL; }
if (table) { table->file->extra(HA_EXTRA_NO_IGNORE_DUP_KEY); table->file->extra(HA_EXTRA_WRITE_CANNOT_REPLACE); table->auto_increment_field_not_null= FALSE; drop_open_table(thd, table, create_table->db, create_table->table_name); table=0; // Safety
} DBUG_VOID_RETURN;}
/*****************************************************************************
Instansiate templates*****************************************************************************/
#ifdef HAVE_EXPLICIT_TEMPLATE_INSTANTIATION
template class List_iterator_fast<List_item>;#ifndef EMBEDDED_LIBRARY
template class I_List<Delayed_insert>;template class I_List_iterator<Delayed_insert>;template class I_List<delayed_row>;#endif /* EMBEDDED_LIBRARY */
#endif /* HAVE_EXPLICIT_TEMPLATE_INSTANTIATION */
|
