Mirrors
/
mariadb
mirror of https://github.com/MariaDB/server
3
0
Fork 0
Code Releases Activity
Browse Source

MDEV-12288 Reset DB_TRX_ID when the history is removed, to speed up MVCC 

Let InnoDB purge reset DB_TRX_ID,DB_ROLL_PTR when the history is removed.

[TODO: It appears that the resetting is not taking place as often as
it could be. We should test that a simple INSERT should eventually
cause row_purge_reset_trx_id() to be invoked unless DROP TABLE is
invoked soon enough.]

The InnoDB clustered index record system columns DB_TRX_ID,DB_ROLL_PTR
are used by multi-versioning. After the history is no longer needed, these
columns can safely be reset to 0 and 1<<55 (to indicate a fresh insert).

When a reader sees 0 in the DB_TRX_ID column, it can instantly determine
that the record is present the read view. There is no need to acquire
the transaction system mutex to check if the transaction exists, because
writes can never be conducted by a transaction whose ID is 0.

The persistent InnoDB undo log used to be split into two parts:
insert_undo and update_undo. The insert_undo log was discarded at
transaction commit or rollback, and the update_undo log was processed
by the purge subsystem. As part of this change, we will only generate
a single undo log for new transactions, and the purge subsystem will
reset the DB_TRX_ID whenever a clustered index record is touched.
That is, all persistent undo log will be preserved at transaction commit
or rollback, to be removed by purge.

The InnoDB redo log format is changed in two ways:
We remove the redo log record type MLOG_UNDO_HDR_REUSE, and
we introduce the MLOG_ZIP_WRITE_TRX_ID record for updating the
DB_TRX_ID,DB_ROLL_PTR in a ROW_FORMAT=COMPRESSED table.

This is also changing the format of persistent InnoDB data files:
undo log and clustered index leaf page records. It will still be
possible via import and export to exchange data files with earlier
versions of MariaDB. The change to clustered index leaf page records
is simple: we allow DB_TRX_ID to be 0.

When it comes to the undo log, we must be able to upgrade from earlier
MariaDB versions after a clean shutdown (no redo log to apply).
While it would be nice to perform a slow shutdown (innodb_fast_shutdown=0)
before an upgrade, to empty the undo logs, we cannot assume that this
has been done. So, separate insert_undo log may exist for recovered
uncommitted transactions. These transactions may be automatically
rolled back, or they may be in XA PREPARE state, in which case InnoDB
will preserve the transaction until an explicit XA COMMIT or XA ROLLBACK.

Upgrade has been tested by starting up MariaDB 10.2 with
./mysql-test-run --manual-gdb innodb.read_only_recovery
and then starting up this patched server with
and without --innodb-read-only.

trx_undo_ptr_t::undo: Renamed from update_undo.

trx_undo_ptr_t::old_insert: Renamed from insert_undo.

trx_rseg_t::undo_list: Renamed from update_undo_list.

trx_rseg_t::undo_cached: Merged from update_undo_cached
and insert_undo_cached.

trx_rseg_t::old_insert_list: Renamed from insert_undo_list.

row_purge_reset_trx_id(): New function to reset the columns.
This will be called for all undo processing in purge
that does not remove the clustered index record.

trx_undo_update_rec_get_update(): Allow trx_id=0 when copying the
old DB_TRX_ID of the record to the undo log.

ReadView::changes_visible(): Allow id==0. (Return true for it.
This is what speeds up the MVCC.)

row_vers_impl_x_locked_low(), row_vers_build_for_semi_consistent_read():
Implement a fast path for DB_TRX_ID=0.

Always initialize the TRX_UNDO_PAGE_TYPE to 0. Remove undo->type.

MLOG_UNDO_HDR_REUSE: Remove. This changes the redo log format!

innobase_start_or_create_for_mysql(): Set srv_undo_sources before
starting any transactions.

The parsing of the MLOG_ZIP_WRITE_TRX_ID record was successfully
tested by running the following:
./mtr --parallel=auto --mysqld=--debug=d,ib_log innodb_zip.bug56680
grep MLOG_ZIP_WRITE_TRX_ID var/*/log/mysqld.1.err
pull/327/merge
Marko Mäkelä 9 years ago
parent
bae0844f65
commit
3c09f148f3
25 changed files with 514 additions and 603 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 3
      mysql-test/suite/innodb_zip/r/bug56680.result
  2. 16
      mysql-test/suite/innodb_zip/t/bug56680.test
  3. 3
      mysql-test/suite/innodb_zip/t/recover.test
  4. 14
      storage/innobase/buf/buf0buf.cc
  5. 9
      storage/innobase/include/mtr0types.h
  6. 41
      storage/innobase/include/page0zip.h
  7. 2
      storage/innobase/include/read0types.h
  8. 20
      storage/innobase/include/trx0rseg.h
  9. 19
      storage/innobase/include/trx0trx.h
  10. 23
      storage/innobase/include/trx0undo.h
  11. 2
      storage/innobase/log/log0log.cc
  12. 16
      storage/innobase/log/log0recv.cc
  13. 101
      storage/innobase/page/page0zip.cc
  14. 7
      storage/innobase/row/row0import.cc
  15. 124
      storage/innobase/row/row0purge.cc
  16. 7
      storage/innobase/row/row0trunc.cc
  17. 3
      storage/innobase/row/row0upd.cc
  18. 17
      storage/innobase/row/row0vers.cc
  19. 33
      storage/innobase/srv/srv0start.cc
  20. 16
      storage/innobase/trx/trx0purge.cc
  21. 41
      storage/innobase/trx/trx0rec.cc
  22. 54
      storage/innobase/trx/trx0roll.cc
  23. 28
      storage/innobase/trx/trx0rseg.cc
  24. 128
      storage/innobase/trx/trx0trx.cc
  25. 390
      storage/innobase/trx/trx0undo.cc

3
mysql-test/suite/innodb_zip/r/bug56680.result
View File

@ -116,5 +116,8 @@ DF
CHECK TABLE bug56680_2;
Table Op Msg_type Msg_text
test.bug56680_2 check status OK
CHECK TABLE bug56680_2;
Table Op Msg_type Msg_text
test.bug56680_2 check status OK
DROP TABLE bug56680_2;
DROP TABLE bug56680;

16
mysql-test/suite/innodb_zip/t/bug56680.test
View File

@ -1,20 +1,18 @@
#Want to skip this test from daily Valgrind execution
--source include/no_valgrind_without_big.inc
# Embedded server tests do not support restarting.
--source include/not_embedded.inc
#
# Bug #56680 InnoDB may return wrong results from a case-insensitive index
#
-- source include/innodb_page_size_small.inc
-- disable_query_log
SET @tx_isolation_orig = @@tx_isolation;
SET @innodb_file_per_table_orig = @@innodb_file_per_table;
# The flag innodb_change_buffering_debug is only available in debug builds.
# It instructs InnoDB to try to evict pages from the buffer pool when
# change buffering is possible, so that the change buffer will be used
# whenever possible.
-- error 0,ER_UNKNOWN_SYSTEM_VARIABLE
SET @innodb_change_buffering_debug_orig = @@innodb_change_buffering_debug;
-- error 0,ER_UNKNOWN_SYSTEM_VARIABLE
SET GLOBAL innodb_change_buffering_debug = 1;
-- enable_query_log
SET GLOBAL tx_isolation='REPEATABLE-READ';
@ -129,11 +127,9 @@ INSERT INTO bug56680_2 VALUES(1,_latin1 0xdf,1);
SELECT HEX(b) FROM bug56680_2 LIMIT 3;
CHECK TABLE bug56680_2;
--let $shutdown_timeout=1
--source include/restart_mysqld.inc
CHECK TABLE bug56680_2;
DROP TABLE bug56680_2;
DROP TABLE bug56680;
-- disable_query_log
SET GLOBAL tx_isolation = @tx_isolation_orig;
SET GLOBAL innodb_file_per_table = @innodb_file_per_table_orig;
-- error 0, ER_UNKNOWN_SYSTEM_VARIABLE
SET GLOBAL innodb_change_buffering_debug = @innodb_change_buffering_debug_orig;

3
mysql-test/suite/innodb_zip/t/recover.test
View File

@ -1,5 +1,4 @@
--source include/have_innodb.inc
--source include/have_innodb_max_16k.inc
--source include/innodb_page_size_small.inc
--source include/not_embedded.inc
--disable_query_log

14
storage/innobase/buf/buf0buf.cc
View File

@ -1284,17 +1284,6 @@ buf_page_print(
read_buf + FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID);
}
if (mach_read_from_2(read_buf + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_TYPE)
== TRX_UNDO_INSERT) {
fprintf(stderr,
"InnoDB: Page may be an insert undo log page\n");
} else if (mach_read_from_2(read_buf + TRX_UNDO_PAGE_HDR
+ TRX_UNDO_PAGE_TYPE)
== TRX_UNDO_UPDATE) {
fprintf(stderr,
"InnoDB: Page may be an update undo log page\n");
}
switch (fil_page_get_type(read_buf)) {
index_id_t index_id;
case FIL_PAGE_INDEX:
@ -1311,6 +1300,9 @@ buf_page_print(
<< " in table " << index->table->name;
}
break;
case FIL_PAGE_UNDO_LOG:
fputs("InnoDB: Page may be an undo log page\n", stderr);
break;
case FIL_PAGE_INODE:
fputs("InnoDB: Page may be an 'inode' page\n", stderr);
break;

9
storage/innobase/include/mtr0types.h
View File

@ -112,9 +112,6 @@ enum mlog_id_t {
/** discard an update undo log header */
MLOG_UNDO_HDR_DISCARD = 23,
/** reuse an insert undo log header */
MLOG_UNDO_HDR_REUSE = 24,
/** create an undo log header */
MLOG_UNDO_HDR_CREATE = 25,
@ -235,8 +232,12 @@ enum mlog_id_t {
redo log about individual pages */
MLOG_INDEX_LOAD = 61,
/** write DB_TRX_ID,DB_ROLL_PTR to a clustered index leaf page
of a ROW_FORMAT=COMPRESSED table */
MLOG_ZIP_WRITE_TRX_ID = 62,
/** biggest value (used in assertions) */
MLOG_BIGGEST_TYPE = MLOG_INDEX_LOAD,
MLOG_BIGGEST_TYPE = MLOG_ZIP_WRITE_TRX_ID,
/** log record for writing/updating crypt data of
a tablespace */

41
storage/innobase/include/page0zip.h
View File

@ -340,18 +340,39 @@ page_zip_write_node_ptr(
ulint ptr, /*!< in: node pointer */
mtr_t* mtr); /*!< in: mini-transaction, or NULL */
/**********************************************************************//**
Write the trx_id and roll_ptr of a record on a B-tree leaf node page. */
/** Write the DB_TRX_ID,DB_ROLL_PTR into a clustered index leaf page record.
@param[in,out] page_zip compressed page
@param[in,out] rec record
@param[in] offsets rec_get_offsets(rec, index)
@param[in] trx_id_field field number of DB_TRX_ID (number of PK fields)
@param[in] trx_id DB_TRX_ID value (transaction identifier)
@param[in] roll_ptr DB_ROLL_PTR value (undo log pointer)
@param[in,out] mtr mini-transaction, or NULL to skip logging */
void
page_zip_write_trx_id_and_roll_ptr(
/*===============================*/
page_zip_des_t* page_zip,/*!< in/out: compressed page */
byte* rec, /*!< in/out: record */
const ulint* offsets,/*!< in: rec_get_offsets(rec, index) */
ulint trx_id_col,/*!< in: column number of TRX_ID in rec */
trx_id_t trx_id, /*!< in: transaction identifier */
roll_ptr_t roll_ptr)/*!< in: roll_ptr */
MY_ATTRIBUTE((nonnull));
page_zip_des_t* page_zip,
byte* rec,
const ulint* offsets,
ulint trx_id_col,
trx_id_t trx_id,
roll_ptr_t roll_ptr,
mtr_t* mtr = NULL)
MY_ATTRIBUTE((nonnull(1,2,3)));
/** Parse a MLOG_ZIP_WRITE_TRX_ID record.
@param[in] ptr redo log buffer
@param[in] end_ptr end of redo log buffer
@param[in,out] page uncompressed page
@param[in,out] page_zip compressed page
@return end of log record
@retval NULL if the log record is incomplete */
byte*
page_zip_parse_write_trx_id(
byte* ptr,
byte* end_ptr,
page_t* page,
page_zip_des_t* page_zip)
MY_ATTRIBUTE((nonnull(1,2), warn_unused_result));
/**********************************************************************//**
Write the "deleted" flag of a record on a compressed page. The flag must

2
storage/innobase/include/read0types.h
View File

@ -163,8 +163,6 @@ public:
const table_name_t& name) const
MY_ATTRIBUTE((warn_unused_result))
{
ut_ad(id > 0);
if (id < m_up_limit_id || id == m_creator_trx_id) {
return(true);

20
storage/innobase/include/trx0rseg.h
View File

@ -162,20 +162,16 @@ struct trx_rseg_t {
ulint curr_size;
/*--------------------------------------------------------*/
/* Fields for update undo logs */
/** List of update undo logs */
UT_LIST_BASE_NODE_T(trx_undo_t) update_undo_list;
/* Fields for undo logs */
/** List of undo logs */
UT_LIST_BASE_NODE_T(trx_undo_t) undo_list;
/** List of update undo log segments cached for fast reuse */
UT_LIST_BASE_NODE_T(trx_undo_t) update_undo_cached;
/** List of undo log segments cached for fast reuse */
UT_LIST_BASE_NODE_T(trx_undo_t) undo_cached;
/*--------------------------------------------------------*/
/* Fields for insert undo logs */
/** List of insert undo logs */
UT_LIST_BASE_NODE_T(trx_undo_t) insert_undo_list;
/** List of insert undo log segments cached for fast reuse */
UT_LIST_BASE_NODE_T(trx_undo_t) insert_undo_cached;
/** List of recovered old insert_undo logs of incomplete
transactions (to roll back or XA COMMIT & purge) */
UT_LIST_BASE_NODE_T(trx_undo_t) old_insert_list;
/*--------------------------------------------------------*/

19
storage/innobase/include/trx0trx.h
View File

@ -842,10 +842,12 @@ struct trx_undo_ptr_t {
trx_rseg_t* rseg; /*!< rollback segment assigned to the
transaction, or NULL if not assigned
yet */
trx_undo_t* insert_undo; /*!< pointer to the insert undo log, or
NULL if no inserts performed yet */
trx_undo_t* update_undo; /*!< pointer to the update undo log, or
NULL if no update performed yet */
trx_undo_t* undo; /*!< pointer to the undo log, or
NULL if nothing logged yet */
trx_undo_t* old_insert; /*!< pointer to recovered
insert undo log, or NULL if no
INSERT transactions were
recovered from old-format undo logs */
};
/** An instance of temporary rollback segment. */
@ -1274,7 +1276,7 @@ struct trx_t {
/** @return whether any persistent undo log has been generated */
bool has_logged_persistent() const
{
return(rsegs.m_redo.insert_undo || rsegs.m_redo.update_undo);
return(rsegs.m_redo.undo);
}
/** @return whether any undo log has been generated */
@ -1283,6 +1285,13 @@ struct trx_t {
return(has_logged_persistent() || rsegs.m_noredo.undo);
}
/** @return whether any undo log has been generated or
recovered */
bool has_logged_or_recovered() const
{
return(has_logged() || rsegs.m_redo.old_insert);
}
/** @return rollback segment for modifying temporary tables */
trx_rseg_t* get_temp_rseg()
{

23
storage/innobase/include/trx0undo.h
View File

@ -257,18 +257,13 @@ A new undo log is created or a cached undo log reused.
@param[in,out] trx transaction
@param[in] rseg rollback segment
@param[out] undo the undo log
@param[in] type TRX_UNDO_INSERT or TRX_UNDO_UPDATE
@retval DB_SUCCESS on success
@retval DB_TOO_MANY_CONCURRENT_TRXS
@retval DB_OUT_OF_FILE_SPACE
@retval DB_READ_ONLY
@retval DB_OUT_OF_MEMORY */
dberr_t
trx_undo_assign_undo(
trx_t* trx,
trx_rseg_t* rseg,
trx_undo_t** undo,
ulint type)
trx_undo_assign_undo(trx_t* trx, trx_rseg_t* rseg, trx_undo_t** undo)
MY_ATTRIBUTE((nonnull, warn_unused_result));
/******************************************************************//**
Sets the state of the undo log segment at a transaction finish.
@ -281,7 +276,7 @@ trx_undo_set_state_at_finish(
/** Set the state of the undo log segment at a XA PREPARE or XA ROLLBACK.
@param[in,out] trx transaction
@param[in,out] undo insert_undo or update_undo log
@param[in,out] undo undo log
@param[in] rollback false=XA PREPARE, true=XA ROLLBACK
@param[in,out] mtr mini-transaction
@return undo log segment header page, x-latched */
@ -305,7 +300,7 @@ trx_undo_update_cleanup(
x-latched */
mtr_t* mtr); /*!< in: mtr */
/** Free an insert or temporary undo log after commit or rollback.
/** Free an old insert or temporary undo log after commit or rollback.
The information is not needed after a commit or rollback, therefore
the data can be discarded.
@param[in,out] undo undo log
@ -343,8 +338,7 @@ trx_undo_parse_page_init(
const byte* end_ptr,/*!< in: buffer end */
page_t* page, /*!< in: page or NULL */
mtr_t* mtr); /*!< in: mtr or NULL */
/** Parse the redo log entry of an undo log page header create or reuse.
@param[in] type MLOG_UNDO_HDR_CREATE or MLOG_UNDO_HDR_REUSE
/** Parse the redo log entry of an undo log page header create.
@param[in] ptr redo log record
@param[in] end_ptr end of log buffer
@param[in,out] page page frame or NULL
@ -352,7 +346,6 @@ trx_undo_parse_page_init(
@return end of log record or NULL */
byte*
trx_undo_parse_page_header(
mlog_id_t type,
const byte* ptr,
const byte* end_ptr,
page_t* page,
@ -402,8 +395,6 @@ struct trx_undo_t {
/*-----------------------------*/
ulint id; /*!< undo log slot number within the
rollback segment */
ulint type; /*!< TRX_UNDO_INSERT or
TRX_UNDO_UPDATE */
ulint state; /*!< state of the corresponding undo log
segment */
ibool del_marks; /*!< relevant only in an update undo
@ -460,8 +451,8 @@ struct trx_undo_t {
/*-------------------------------------------------------------*/
/** Transaction undo log page header offsets */
/* @{ */
#define TRX_UNDO_PAGE_TYPE 0 /*!< TRX_UNDO_INSERT or
TRX_UNDO_UPDATE */
#define TRX_UNDO_PAGE_TYPE 0 /*!< unused; 0 (before MariaDB 10.3.1:
TRX_UNDO_INSERT or TRX_UNDO_UPDATE) */
#define TRX_UNDO_PAGE_START 2 /*!< Byte offset where the undo log
records for the LATEST transaction
start on this page (remember that
@ -559,7 +550,7 @@ page of an update undo log segment. */
#define TRX_UNDO_LOG_OLD_HDR_SIZE (34 + FLST_NODE_SIZE)
/* Note: the writing of the undo log old header is coded by a log record
MLOG_UNDO_HDR_CREATE or MLOG_UNDO_HDR_REUSE. The appending of an XID to the
MLOG_UNDO_HDR_CREATE. The appending of an XID to the
header is logged separately. In this sense, the XID is not really a member
of the undo log header. TODO: do not append the XID to the log header if XA
is not needed by the user. The XID wastes about 150 bytes of space in every

2
storage/innobase/log/log0log.cc
View File

@ -68,7 +68,7 @@ c-function and its parameters are written to the log to
reduce the size of the log.
3a) You should not add parameters to these kind of functions
(e.g. trx_undo_header_create(), trx_undo_insert_header_reuse())
(e.g. trx_undo_header_create())
3b) You should not add such functionality which either change
working when compared with the old or are dependent on data

16
storage/innobase/log/log0recv.cc
View File

@ -1424,10 +1424,8 @@ parse_log:
ptr = trx_undo_parse_discard_latest(ptr, end_ptr, page, mtr);
break;
case MLOG_UNDO_HDR_CREATE:
case MLOG_UNDO_HDR_REUSE:
ut_ad(!page || page_type == FIL_PAGE_UNDO_LOG);
ptr = trx_undo_parse_page_header(type, ptr, end_ptr,
page, mtr);
ptr = trx_undo_parse_page_header(ptr, end_ptr, page, mtr);
break;
case MLOG_REC_MIN_MARK: case MLOG_COMP_REC_MIN_MARK:
ut_ad(!page || fil_page_type_is_index(page_type));
@ -1496,6 +1494,12 @@ parse_log:
ptr, end_ptr, page, page_zip, index);
}
break;
case MLOG_ZIP_WRITE_TRX_ID:
/* This must be a clustered index leaf page. */
ut_ad(!page || page_type == FIL_PAGE_INDEX);
ptr = page_zip_parse_write_trx_id(ptr, end_ptr,
page, page_zip);
break;
case MLOG_FILE_WRITE_CRYPT_DATA:
dberr_t err;
ptr = const_cast<byte*>(fil_parse_write_crypt_data(ptr, end_ptr, block, &err));
@ -3654,9 +3658,6 @@ get_mlog_string(mlog_id_t type)
case MLOG_UNDO_HDR_DISCARD:
return("MLOG_UNDO_HDR_DISCARD");
case MLOG_UNDO_HDR_REUSE:
return("MLOG_UNDO_HDR_REUSE");
case MLOG_UNDO_HDR_CREATE:
return("MLOG_UNDO_HDR_CREATE");
@ -3737,6 +3738,9 @@ get_mlog_string(mlog_id_t type)
case MLOG_ZIP_PAGE_REORGANIZE:
return("MLOG_ZIP_PAGE_REORGANIZE");
case MLOG_ZIP_WRITE_TRX_ID:
return("MLOG_ZIP_WRITE_TRX_ID");
case MLOG_FILE_RENAME2:
return("MLOG_FILE_RENAME2");

101
storage/innobase/page/page0zip.cc
View File

@ -4140,17 +4140,23 @@ page_zip_write_node_ptr(
}
}
/**********************************************************************//**
Write the trx_id and roll_ptr of a record on a B-tree leaf node page. */
/** Write the DB_TRX_ID,DB_ROLL_PTR into a clustered index leaf page record.
@param[in,out] page_zip compressed page
@param[in,out] rec record
@param[in] offsets rec_get_offsets(rec, index)
@param[in] trx_id_field field number of DB_TRX_ID (number of PK fields)
@param[in] trx_id DB_TRX_ID value (transaction identifier)
@param[in] roll_ptr DB_ROLL_PTR value (undo log pointer)
@param[in,out] mtr mini-transaction, or NULL to skip logging */
void
page_zip_write_trx_id_and_roll_ptr(
/*===============================*/
page_zip_des_t* page_zip,/*!< in/out: compressed page */
byte* rec, /*!< in/out: record */
const ulint* offsets,/*!< in: rec_get_offsets(rec, index) */
ulint trx_id_col,/*!< in: column number of TRX_ID in rec */
trx_id_t trx_id, /*!< in: transaction identifier */
roll_ptr_t roll_ptr)/*!< in: roll_ptr */
page_zip_des_t* page_zip,
byte* rec,
const ulint* offsets,
ulint trx_id_col,
trx_id_t trx_id,
roll_ptr_t roll_ptr,
mtr_t* mtr)
{
byte* field;
byte* storage;
@ -4202,6 +4208,83 @@ page_zip_write_trx_id_and_roll_ptr(
UNIV_MEM_ASSERT_RW(rec - rec_offs_extra_size(offsets),
rec_offs_extra_size(offsets));
UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));
if (mtr) {
byte* log_ptr = mlog_open(
mtr, 11 + 2 + 2 + DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
if (UNIV_UNLIKELY(!log_ptr)) {
return;
}
log_ptr = mlog_write_initial_log_record_fast(
(byte*) field, MLOG_ZIP_WRITE_TRX_ID, log_ptr, mtr);
mach_write_to_2(log_ptr, page_offset(field));
log_ptr += 2;
mach_write_to_2(log_ptr, storage - page_zip->data);
log_ptr += 2;
memcpy(log_ptr, field, DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
log_ptr += DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN;
mlog_close(mtr, log_ptr);
}
}
/** Parse a MLOG_ZIP_WRITE_TRX_ID record.
@param[in] ptr redo log buffer
@param[in] end_ptr end of redo log buffer
@param[in,out] page uncompressed page
@param[in,out] page_zip compressed page
@return end of log record
@retval NULL if the log record is incomplete */
byte*
page_zip_parse_write_trx_id(
byte* ptr,
byte* end_ptr,
page_t* page,
page_zip_des_t* page_zip)
{
byte* const end = 2 + 2 + DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN + ptr;
if (UNIV_UNLIKELY(end_ptr < end)) {
return(NULL);
}
uint offset = mach_read_from_2(ptr);
uint z_offset = mach_read_from_2(ptr + 2);
if (offset < PAGE_ZIP_START
|| offset >= UNIV_PAGE_SIZE
|| z_offset >= UNIV_PAGE_SIZE) {
corrupt:
recv_sys->found_corrupt_log = TRUE;
return(NULL);
}
if (page) {
if (!page_zip || !page_is_leaf(page)) {
goto corrupt;
}
#ifdef UNIV_ZIP_DEBUG
ut_a(page_zip_validate(page_zip, page, NULL));
#endif /* UNIV_ZIP_DEBUG */
byte* field = page + offset;
byte* storage = page_zip->data + z_offset;
if (storage >= page_zip_dir_start(page_zip)) {
goto corrupt;
}
memcpy(field, ptr + 4, DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
memcpy(storage, ptr + 4, DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
#ifdef UNIV_ZIP_DEBUG
ut_a(page_zip_validate(page_zip, page, NULL));
#endif /* UNIV_ZIP_DEBUG */
}
return end;
}
/**********************************************************************//**

7
storage/innobase/row/row0import.cc
View File

@ -3380,9 +3380,8 @@ row_import_for_mysql(
mutex_enter(&trx->undo_mutex);
/* TODO: Do not write any undo log for the IMPORT cleanup. */
trx_undo_t** pundo = &trx->rsegs.m_redo.update_undo;
err = trx_undo_assign_undo(trx, trx->rsegs.m_redo.rseg, pundo,
TRX_UNDO_UPDATE);
err = trx_undo_assign_undo(trx, trx->rsegs.m_redo.rseg,
&trx->rsegs.m_redo.undo);
mutex_exit(&trx->undo_mutex);
@ -3393,7 +3392,7 @@ row_import_for_mysql(
return(row_import_cleanup(prebuilt, trx, err));
} else if (trx->rsegs.m_redo.update_undo == 0) {
} else if (trx->rsegs.m_redo.undo == 0) {
err = DB_TOO_MANY_CONCURRENT_TRXS;
return(row_import_cleanup(prebuilt, trx, err));

124
storage/innobase/row/row0purge.cc
View File

@ -663,6 +663,75 @@ row_purge_del_mark(
return(row_purge_remove_clust_if_poss(node));
}
/** Reset DB_TRX_ID, DB_ROLL_PTR of a clustered index record
whose old history can no longer be observed.
@param[in,out] node purge node
@param[in,out] mtr mini-transaction (will be started and committed) */
static
void
row_purge_reset_trx_id(purge_node_t* node, mtr_t* mtr)
{
ut_ad(rw_lock_own(dict_operation_lock, RW_LOCK_S));
/* Reset DB_TRX_ID, DB_ROLL_PTR for old records. */
mtr->start();
if (row_purge_reposition_pcur(BTR_MODIFY_LEAF, node, mtr)) {
dict_index_t* index = dict_table_get_first_index(
node->table);
ulint trx_id_pos = index->n_uniq ? index->n_uniq : 1;
rec_t* rec = btr_pcur_get_rec(&node->pcur);
mem_heap_t* heap = NULL;
/* Reserve enough offsets for the PRIMARY KEY and 2 columns
so that we can access DB_TRX_ID, DB_ROLL_PTR. */
ulint offsets_[REC_OFFS_HEADER_SIZE + MAX_REF_PARTS + 2];
rec_offs_init(offsets_);
ulint* offsets = rec_get_offsets(
rec, index, offsets_, trx_id_pos + 2, &heap);
ut_ad(heap == NULL);
ut_ad(dict_index_get_nth_field(index, trx_id_pos)
->col->mtype == DATA_SYS);
ut_ad(dict_index_get_nth_field(index, trx_id_pos)
->col->prtype == (DATA_TRX_ID | DATA_NOT_NULL));
ut_ad(dict_index_get_nth_field(index, trx_id_pos + 1)
->col->mtype == DATA_SYS);
ut_ad(dict_index_get_nth_field(index, trx_id_pos + 1)
->col->prtype == (DATA_ROLL_PTR | DATA_NOT_NULL));
/* Only update the record if DB_ROLL_PTR matches (the
record has not been modified after this transaction
became purgeable) */
if (node->roll_ptr
== row_get_rec_roll_ptr(rec, index, offsets)) {
ut_ad(!rec_get_deleted_flag(rec,
rec_offs_comp(offsets)));
mtr->set_named_space(index->space);
if (page_zip_des_t* page_zip
= buf_block_get_page_zip(
btr_pcur_get_block(&node->pcur))) {
page_zip_write_trx_id_and_roll_ptr(
page_zip, rec, offsets, trx_id_pos,
0, 1ULL << ROLL_PTR_INSERT_FLAG_POS,
mtr);
} else {
ulint len;
byte* ptr = rec_get_nth_field(
rec, offsets, trx_id_pos, &len);
ut_ad(len == DATA_TRX_ID_LEN);
memset(ptr, 0, DATA_TRX_ID_LEN
+ DATA_ROLL_PTR_LEN);
ptr[DATA_TRX_ID_LEN] = 1U
<< (ROLL_PTR_INSERT_FLAG_POS - CHAR_BIT
* (DATA_ROLL_PTR_LEN - 1));
mlog_log_string(ptr, DATA_TRX_ID_LEN
+ DATA_ROLL_PTR_LEN, mtr);
}
}
}
mtr->commit();
}
/***********************************************************//**
Purges an update of an existing record. Also purges an update of a delete
marked record if that record contained an externally stored field. */
@ -713,6 +782,8 @@ row_purge_upd_exist_or_extern_func(
mem_heap_free(heap);
skip_secondaries:
mtr_t mtr;
dict_index_t* index = dict_table_get_first_index(node->table);
/* Free possible externally stored fields */
for (ulint i = 0; i < upd_get_n_fields(node->update); i++) {
@ -724,12 +795,10 @@ skip_secondaries:
buf_block_t* block;
ulint internal_offset;
byte* data_field;
dict_index_t* index;
ibool is_insert;
ulint rseg_id;
ulint page_no;
ulint offset;
mtr_t mtr;
/* We use the fact that new_val points to
undo_rec and get thus the offset of
@ -759,7 +828,6 @@ skip_secondaries:
/* We have to acquire an SX-latch to the clustered
index tree (exclude other tree changes) */
index = dict_table_get_first_index(node->table);
mtr_sx_lock(dict_index_get_lock(index), &mtr);
mtr.set_named_space(index->space);
@ -794,6 +862,8 @@ skip_secondaries:
mtr_commit(&mtr);
}
}
row_purge_reset_trx_id(node, &mtr);
}
#ifdef UNIV_DEBUG
@ -819,10 +889,8 @@ row_purge_parse_undo_rec(
{
dict_index_t* clust_index;
byte* ptr;
trx_t* trx;
undo_no_t undo_no;
table_id_t table_id;
trx_id_t trx_id;
roll_ptr_t roll_ptr;
ulint info_bits;
ulint type;
@ -836,16 +904,22 @@ row_purge_parse_undo_rec(
node->rec_type = type;
if (type == TRX_UNDO_UPD_DEL_REC && !*updated_extern) {
return(false);
switch (type) {
case TRX_UNDO_INSERT_REC:
break;
default:
#ifdef UNIV_DEBUG
ut_ad(0);
return false;
case TRX_UNDO_UPD_DEL_REC:
case TRX_UNDO_UPD_EXIST_REC:
case TRX_UNDO_DEL_MARK_REC:
#endif /* UNIV_DEBUG */
ptr = trx_undo_update_rec_get_sys_cols(ptr, &node->trx_id,
&roll_ptr, &info_bits);
break;
}
ptr = trx_undo_update_rec_get_sys_cols(ptr, &trx_id, &roll_ptr,
&info_bits);
node->table = NULL;
node->trx_id = trx_id;
/* Prevent DROP TABLE etc. from running when we are doing the purge
for this row */
@ -868,7 +942,8 @@ try_again:
goto err_exit;
}
if (node->table->n_v_cols && !node->table->vc_templ
if (type != TRX_UNDO_INSERT_REC
&& node->table->n_v_cols && !node->table->vc_templ
&& dict_table_has_indexed_v_cols(node->table)) {
/* Need server fully up for virtual column computation */
if (!mysqld_server_started) {
@ -893,28 +968,23 @@ try_again:
/* The table was corrupt in the data dictionary.
dict_set_corrupted() works on an index, and
we do not have an index to call it with. */
close_exit:
dict_table_close(node->table, FALSE, FALSE);
err_exit:
rw_lock_s_unlock(dict_operation_lock);
return(false);
}
if (type == TRX_UNDO_UPD_EXIST_REC
&& (node->cmpl_info & UPD_NODE_NO_ORD_CHANGE)
&& !*updated_extern) {
/* Purge requires no changes to indexes: we may return */
goto close_exit;
}
ptr = trx_undo_rec_get_row_ref(ptr, clust_index, &(node->ref),
node->heap);
trx = thr_get_trx(thr);
if (type == TRX_UNDO_INSERT_REC) {
return(true);
}
ptr = trx_undo_update_rec_get_update(ptr, clust_index, type, trx_id,
roll_ptr, info_bits, trx,
ptr = trx_undo_update_rec_get_update(ptr, clust_index, type,
node->trx_id,
roll_ptr, info_bits,
thr_get_trx(thr),
node->heap, &(node->update));
/* Read to the partial row the fields that occur in indexes */
@ -967,6 +1037,8 @@ row_purge_record_func(
break;
default:
if (!updated_extern) {
mtr_t mtr;
row_purge_reset_trx_id(node, &mtr);
break;
}
/* fall through */

7
storage/innobase/row/row0trunc.cc
View File

@ -1841,11 +1841,8 @@ row_truncate_table_for_mysql(
till some point. Associate rollback segment to record undo log. */
if (!dict_table_is_temporary(table)) {
mutex_enter(&trx->undo_mutex);
trx_undo_t** pundo = &trx->rsegs.m_redo.update_undo;
err = trx_undo_assign_undo(
trx, trx->rsegs.m_redo.rseg, pundo, TRX_UNDO_UPDATE);
err = trx_undo_assign_undo(trx, trx->rsegs.m_redo.rseg,
&trx->rsegs.m_redo.undo);
mutex_exit(&trx->undo_mutex);
DBUG_EXECUTE_IF("ib_err_trunc_assigning_undo_log",

3
storage/innobase/row/row0upd.cc
View File

@ -2582,8 +2582,7 @@ row_upd_clust_rec_by_insert_inherit_func(
data[BTR_EXTERN_LEN] &= ~BTR_EXTERN_OWNER_FLAG;
data[BTR_EXTERN_LEN] |= BTR_EXTERN_INHERITED_FLAG;
/* The BTR_EXTERN_INHERITED_FLAG only matters in
rollback of a fresh insert (insert_undo log).
Purge (operating on update_undo log) will always free
rollback of a fresh insert. Purge will always free
the extern fields of a delete-marked row. */
inherit = true;

17
storage/innobase/row/row0vers.cc
View File

@ -97,12 +97,25 @@ row_vers_impl_x_locked_low(
ut_ad(rec_offs_validate(rec, index, offsets));
if (ulint trx_id_offset = clust_index->trx_id_offset) {
trx_id = mach_read_from_6(clust_rec + trx_id_offset);
if (trx_id == 0) {
/* The transaction history was already purged. */
DBUG_RETURN(0);
}
}
heap = mem_heap_create(1024);
clust_offsets = rec_get_offsets(
clust_rec, clust_index, NULL, ULINT_UNDEFINED, &heap);
trx_id = row_get_rec_trx_id(clust_rec, clust_index, clust_offsets);
if (trx_id == 0) {
/* The transaction history was already purged. */
mem_heap_free(heap);
DBUG_RETURN(0);
}
corrupt = FALSE;
trx_t* trx = trx_rw_is_active(trx_id, &corrupt, true);
@ -1262,6 +1275,10 @@ row_vers_build_for_semi_consistent_read(
rec_trx_id = version_trx_id;
}
if (!version_trx_id) {
goto committed_version_trx;
}
trx_sys_mutex_enter();
version_trx = trx_get_rw_trx_by_id(version_trx_id);
/* Because version_trx is a read-write transaction,

33
storage/innobase/srv/srv0start.cc
View File

@ -2543,6 +2543,20 @@ files_checked:
thread_started[4 + SRV_MAX_N_IO_THREADS] = true;
srv_start_state |= SRV_START_STATE_LOCK_SYS
| SRV_START_STATE_MONITOR;
ut_a(trx_purge_state() == PURGE_STATE_INIT);
if (srv_force_recovery < SRV_FORCE_NO_BACKGROUND) {
srv_undo_sources = true;
/* Create the dict stats gathering thread */
srv_dict_stats_thread_active = true;
dict_stats_thread_handle = os_thread_create(
dict_stats_thread, NULL, NULL);
/* Create the thread that will optimize the
FULLTEXT search index subsystem. */
fts_optimize_init();
}
}
/* Create the SYS_FOREIGN and SYS_FOREIGN_COLS system tables */
@ -2576,16 +2590,7 @@ files_checked:
}
trx_temp_rseg_create();
}
srv_is_being_started = false;
ut_a(trx_purge_state() == PURGE_STATE_INIT);
/* Create the master thread which does purge and other utility
operations */
if (!srv_read_only_mode) {
thread_handles[1 + SRV_MAX_N_IO_THREADS] = os_thread_create(
srv_master_thread,
NULL, thread_ids + (1 + SRV_MAX_N_IO_THREADS));
@ -2593,16 +2598,10 @@ files_checked:
srv_start_state_set(SRV_START_STATE_MASTER);
}
srv_is_being_started = false;
if (!srv_read_only_mode && srv_operation == SRV_OPERATION_NORMAL
&& srv_force_recovery < SRV_FORCE_NO_BACKGROUND) {
srv_undo_sources = true;
/* Create the dict stats gathering thread */
srv_dict_stats_thread_active = true;
dict_stats_thread_handle = os_thread_create(
dict_stats_thread, NULL, NULL);
/* Create the thread that will optimize the FTS sub-system. */
fts_optimize_init();
thread_handles[5 + SRV_MAX_N_IO_THREADS] = os_thread_create(
srv_purge_coordinator_thread,

16
storage/innobase/trx/trx0purge.cc
View File

@ -248,7 +248,7 @@ trx_purge_add_update_undo_to_history(
x-latched */
mtr_t* mtr) /*!< in: mtr */
{
trx_undo_t* undo = trx->rsegs.m_redo.update_undo;
trx_undo_t* undo = trx->rsegs.m_redo.undo;
trx_rseg_t* rseg = undo->rseg;
trx_rsegf_t* rseg_header = trx_rsegf_get(
rseg->space, rseg->page_no, mtr);
@ -954,19 +954,7 @@ trx_purge_initiate_truncate(
ulint cached_undo_size = 0;
for (trx_undo_t* undo =
UT_LIST_GET_FIRST(rseg->update_undo_cached);
undo != NULL && all_free;
undo = UT_LIST_GET_NEXT(undo_list, undo)) {
if (limit->trx_no < undo->trx_id) {
all_free = false;
} else {
cached_undo_size += undo->size;
}
}
for (trx_undo_t* undo =
UT_LIST_GET_FIRST(rseg->insert_undo_cached);
UT_LIST_GET_FIRST(rseg->undo_cached);
undo != NULL && all_free;
undo = UT_LIST_GET_NEXT(undo_list, undo)) {

41
storage/innobase/trx/trx0rec.cc
View File

@ -469,8 +469,8 @@ trx_undo_page_report_insert(
ulint i;
ut_ad(dict_index_is_clust(index));
ut_ad(mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
+ TRX_UNDO_PAGE_TYPE) == TRX_UNDO_INSERT);
ut_ad(*reinterpret_cast<uint16*>(TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_TYPE
+ undo_page) == 0);
first_free = mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
+ TRX_UNDO_PAGE_FREE);
@ -875,13 +875,10 @@ trx_undo_page_report_modify(
ut_a(dict_index_is_clust(index));
ut_ad(rec_offs_validate(rec, index, offsets));
ut_ad(mach_read_from_2(TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_TYPE
+ undo_page) == TRX_UNDO_UPDATE
|| (dict_table_is_temporary(table)
&& mach_read_from_2(TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_TYPE
+ undo_page) == TRX_UNDO_INSERT));
trx_undo_t* update_undo = dict_table_is_temporary(table)
? NULL : trx->rsegs.m_redo.update_undo;
ut_ad(*reinterpret_cast<uint16*>(TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_TYPE
+ undo_page) == 0);
trx_undo_t* undo = dict_table_is_temporary(table)
? NULL : trx->rsegs.m_redo.undo;
first_free = mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
+ TRX_UNDO_PAGE_FREE);
@ -1109,8 +1106,8 @@ trx_undo_page_report_modify(
/* Notify purge that it eventually has to
free the old externally stored field */
if (update_undo) {
update_undo->del_marks = TRUE;
if (undo) {
undo->del_marks = TRUE;
}
*type_cmpl_ptr |= TRX_UNDO_UPD_EXTERN;
@ -1180,8 +1177,8 @@ trx_undo_page_report_modify(
double mbr[SPDIMS * 2];
mem_heap_t* row_heap = NULL;
if (update_undo) {
update_undo->del_marks = TRUE;
if (undo) {
undo->del_marks = TRUE;
}
if (trx_undo_left(undo_page, ptr) < 5) {
@ -1485,7 +1482,7 @@ trx_undo_update_rec_get_update(
buf = static_cast<byte*>(mem_heap_alloc(heap, DATA_TRX_ID_LEN));
trx_write_trx_id(buf, trx_id);
mach_write_to_6(buf, trx_id);
upd_field_set_field_no(upd_field,
dict_index_get_sys_col_pos(index, DATA_TRX_ID),
@ -1901,23 +1898,13 @@ trx_undo_report_row_operation(
not listed there. */
trx->mod_tables.insert(index->table);
pundo = !rec
? &trx->rsegs.m_redo.insert_undo
: &trx->rsegs.m_redo.update_undo;
pundo = &trx->rsegs.m_redo.undo;
rseg = trx->rsegs.m_redo.rseg;
}
mutex_enter(&trx->undo_mutex);
dberr_t err;
if (*pundo) {
err = DB_SUCCESS;
} else if (!rec || is_temp) {
err = trx_undo_assign_undo(trx, rseg, pundo, TRX_UNDO_INSERT);
} else {
err = trx_undo_assign_undo(trx, rseg, pundo, TRX_UNDO_UPDATE);
}
dberr_t err = *pundo ? DB_SUCCESS : trx_undo_assign_undo(
trx, rseg, pundo);
trx_undo_t* undo = *pundo;
ut_ad((err == DB_SUCCESS) == (undo != NULL));

54
storage/innobase/trx/trx0roll.cc
View File

@ -102,7 +102,7 @@ trx_rollback_to_savepoint_low(
trx->error_state = DB_SUCCESS;
if (trx->has_logged()) {
if (trx->has_logged_or_recovered()) {
ut_ad(trx->rsegs.m_redo.rseg != 0
|| trx->rsegs.m_noredo.rseg != 0);
@ -213,24 +213,27 @@ trx_rollback_low(
case TRX_STATE_PREPARED:
ut_ad(!trx_is_autocommit_non_locking(trx));
if (trx->has_logged_persistent()) {
if (trx->rsegs.m_redo.undo || trx->rsegs.m_redo.old_insert) {
/* Change the undo log state back from
TRX_UNDO_PREPARED to TRX_UNDO_ACTIVE
so that if the system gets killed,
recovery will perform the rollback. */
trx_undo_ptr_t* undo_ptr = &trx->rsegs.m_redo;
ut_ad(!trx->rsegs.m_redo.undo
|| trx->rsegs.m_redo.undo->rseg
== trx->rsegs.m_redo.rseg);
ut_ad(!trx->rsegs.m_redo.old_insert
|| trx->rsegs.m_redo.old_insert->rseg
== trx->rsegs.m_redo.rseg);
mtr_t mtr;
mtr.start();
mutex_enter(&trx->rsegs.m_redo.rseg->mutex);
if (undo_ptr->insert_undo != NULL) {
trx_undo_set_state_at_prepare(
trx, undo_ptr->insert_undo,
true, &mtr);
if (trx_undo_t* undo = trx->rsegs.m_redo.undo) {
trx_undo_set_state_at_prepare(trx, undo, true,
&mtr);
}
if (undo_ptr->update_undo != NULL) {
trx_undo_set_state_at_prepare(
trx, undo_ptr->update_undo,
true, &mtr);
if (trx_undo_t* undo = trx->rsegs.m_redo.old_insert) {
trx_undo_set_state_at_prepare(trx, undo, true,
&mtr);
}
mutex_exit(&trx->rsegs.m_redo.rseg->mutex);
/* Persist the XA ROLLBACK, so that crash
@ -899,20 +902,12 @@ trx_roll_try_truncate(trx_t* trx)
trx->pages_undone = 0;
undo_no_t undo_no = trx->undo_no;
trx_undo_t* insert_undo = trx->rsegs.m_redo.insert_undo;
trx_undo_t* update_undo = trx->rsegs.m_redo.update_undo;
if (insert_undo || update_undo) {
mutex_enter(&trx->rsegs.m_redo.rseg->mutex);
if (insert_undo) {
ut_ad(insert_undo->rseg == trx->rsegs.m_redo.rseg);
trx_undo_truncate_end(insert_undo, undo_no, false);
}
if (update_undo) {
ut_ad(update_undo->rseg == trx->rsegs.m_redo.rseg);
trx_undo_truncate_end(update_undo, undo_no, false);
}
mutex_exit(&trx->rsegs.m_redo.rseg->mutex);
if (trx_undo_t* undo = trx->rsegs.m_redo.undo) {
ut_ad(undo->rseg == trx->rsegs.m_redo.rseg);
mutex_enter(&undo->rseg->mutex);
trx_undo_truncate_end(undo, undo_no, false);
mutex_exit(&undo->rseg->mutex);
}
if (trx_undo_t* undo = trx->rsegs.m_noredo.undo) {
@ -987,8 +982,8 @@ trx_roll_pop_top_rec_of_trx(trx_t* trx, roll_ptr_t* roll_ptr, mem_heap_t* heap)
}
trx_undo_t* undo;
trx_undo_t* insert = trx->rsegs.m_redo.insert_undo;
trx_undo_t* update = trx->rsegs.m_redo.update_undo;
trx_undo_t* insert = trx->rsegs.m_redo.old_insert;
trx_undo_t* update = trx->rsegs.m_redo.undo;
trx_undo_t* temp = trx->rsegs.m_noredo.undo;
const undo_no_t limit = trx->roll_limit;
@ -999,7 +994,8 @@ trx_roll_pop_top_rec_of_trx(trx_t* trx, roll_ptr_t* roll_ptr, mem_heap_t* heap)
ut_ad(!update || !temp || update->empty || temp->empty
|| update->top_undo_no != temp->top_undo_no);
if (insert && !insert->empty && limit <= insert->top_undo_no) {
if (UNIV_LIKELY_NULL(insert)
&& !insert->empty && limit <= insert->top_undo_no) {
if (update && !update->empty
&& update->top_undo_no > insert->top_undo_no) {
undo = update;
@ -1033,7 +1029,7 @@ trx_roll_pop_top_rec_of_trx(trx_t* trx, roll_ptr_t* roll_ptr, mem_heap_t* heap)
trx_undo_rec_t* undo_rec = trx_roll_pop_top_rec(trx, undo, &mtr);
const undo_no_t undo_no = trx_undo_rec_get_undo_no(undo_rec);
if (trx_undo_rec_get_type(undo_rec) == TRX_UNDO_INSERT_REC) {
ut_ad(undo == insert || undo == temp);
ut_ad(undo == insert || undo == update || undo == temp);
*roll_ptr |= 1ULL << ROLL_PTR_INSERT_FLAG_POS;
} else {
ut_ad(undo == update || undo == temp);

28
storage/innobase/trx/trx0rseg.cc
View File

@ -115,29 +115,16 @@ trx_rseg_mem_free(trx_rseg_t* rseg)
mutex_free(&rseg->mutex);
/* There can't be any active transactions. */
ut_a(UT_LIST_GET_LEN(rseg->update_undo_list) == 0);
ut_a(UT_LIST_GET_LEN(rseg->insert_undo_list) == 0);
ut_a(UT_LIST_GET_LEN(rseg->undo_list) == 0);
ut_a(UT_LIST_GET_LEN(rseg->old_insert_list) == 0);
for (undo = UT_LIST_GET_FIRST(rseg->update_undo_cached);
for (undo = UT_LIST_GET_FIRST(rseg->undo_cached);
undo != NULL;
undo = next_undo) {
next_undo = UT_LIST_GET_NEXT(undo_list, undo);
UT_LIST_REMOVE(rseg->update_undo_cached, undo);
MONITOR_DEC(MONITOR_NUM_UNDO_SLOT_CACHED);
trx_undo_mem_free(undo);
}
for (undo = UT_LIST_GET_FIRST(rseg->insert_undo_cached);
undo != NULL;
undo = next_undo) {
next_undo = UT_LIST_GET_NEXT(undo_list, undo);
UT_LIST_REMOVE(rseg->insert_undo_cached, undo);
UT_LIST_REMOVE(rseg->undo_cached, undo);
MONITOR_DEC(MONITOR_NUM_UNDO_SLOT_CACHED);
@ -167,10 +154,9 @@ trx_rseg_mem_create(ulint id, ulint space, ulint page_no)
? LATCH_ID_REDO_RSEG : LATCH_ID_NOREDO_RSEG,
&rseg->mutex);
UT_LIST_INIT(rseg->update_undo_list, &trx_undo_t::undo_list);
UT_LIST_INIT(rseg->update_undo_cached, &trx_undo_t::undo_list);
UT_LIST_INIT(rseg->insert_undo_list, &trx_undo_t::undo_list);
UT_LIST_INIT(rseg->insert_undo_cached, &trx_undo_t::undo_list);
UT_LIST_INIT(rseg->undo_list, &trx_undo_t::undo_list);
UT_LIST_INIT(rseg->old_insert_list, &trx_undo_t::undo_list);
UT_LIST_INIT(rseg->undo_cached, &trx_undo_t::undo_list);
return(rseg);
}

128
storage/innobase/trx/trx0trx.cc
View File

@ -717,8 +717,6 @@ void
trx_resurrect_table_locks(
/*======================*/
trx_t* trx, /*!< in/out: transaction */
const trx_undo_ptr_t* undo_ptr,
/*!< in: pointer to undo segment. */
const trx_undo_t* undo) /*!< in: undo log */
{
mtr_t mtr;
@ -726,8 +724,6 @@ trx_resurrect_table_locks(
trx_undo_rec_t* undo_rec;
table_id_set tables;
ut_ad(undo == undo_ptr->insert_undo || undo == undo_ptr->update_undo);
if (trx_state_eq(trx, TRX_STATE_COMMITTED_IN_MEMORY) || undo->empty) {
return;
@ -785,13 +781,9 @@ trx_resurrect_table_locks(
}
lock_table_ix_resurrect(table, trx);
DBUG_PRINT("ib_trx",
("resurrect" TRX_ID_FMT
" table '%s' IX lock from %s undo",
trx_get_id_for_print(trx),
table->name.m_name,
undo == undo_ptr->insert_undo
? "insert" : "update"));
DBUG_LOG("ib_trx",
"resurrect " << ib::hex(trx->id)
<< " IX lock on " << table->name);
dict_table_close(table, FALSE, FALSE);
}
@ -823,7 +815,7 @@ trx_resurrect_insert(
++trx->rsegs.m_redo.rseg->trx_ref_count;
*trx->xid = undo->xid;
trx->id = undo->trx_id;
trx->rsegs.m_redo.insert_undo = undo;
trx->rsegs.m_redo.old_insert = undo;
trx->is_recovered = true;
/* This is single-threaded startup code, we do not need the
@ -873,14 +865,6 @@ trx_resurrect_insert(
trx->no = TRX_ID_MAX;
}
/* trx_start_low() is not called with resurrect, so need to initialize
start time here.*/
if (trx->state == TRX_STATE_ACTIVE
|| trx->state == TRX_STATE_PREPARED) {
trx->start_time = ut_time();
}
if (undo->dict_operation) {
trx_set_dict_operation(trx, TRX_DICT_OP_TABLE);
trx->table_id = undo->table_id;
@ -952,7 +936,7 @@ trx_resurrect_update(
++trx->rsegs.m_redo.rseg->trx_ref_count;
*trx->xid = undo->xid;
trx->id = undo->trx_id;
trx->rsegs.m_redo.update_undo = undo;
trx->rsegs.m_redo.undo = undo;
trx->is_recovered = true;
/* This is single-threaded startup code, we do not need the
@ -974,13 +958,6 @@ trx_resurrect_update(
trx->no = TRX_ID_MAX;
}
/* trx_start_low() is not called with resurrect, so need to initialize
start time here.*/
if (trx->state == TRX_STATE_ACTIVE
|| trx->state == TRX_STATE_PREPARED) {
trx->start_time = ut_time();
}
if (undo->dict_operation) {
trx_set_dict_operation(trx, TRX_DICT_OP_TABLE);
trx->table_id = undo->table_id;
@ -1009,6 +986,7 @@ trx_lists_init_at_db_start()
/* Look from the rollback segments if there exist undo logs for
transactions. */
const ib_time_t start_time = ut_time();
for (ulint i = 0; i < TRX_SYS_N_RSEGS; ++i) {
trx_undo_t* undo;
@ -1020,23 +998,24 @@ trx_lists_init_at_db_start()
continue;
}
/* Resurrect transactions that were doing inserts. */
for (undo = UT_LIST_GET_FIRST(rseg->insert_undo_list);
/* Resurrect transactions that were doing inserts
using the old separate insert_undo log. */
for (undo = UT_LIST_GET_FIRST(rseg->old_insert_list);
undo != NULL;
undo = UT_LIST_GET_NEXT(undo_list, undo)) {
trx_t* trx;
trx = trx_resurrect_insert(undo, rseg);
trx->start_time = start_time;
trx_sys_rw_trx_add(trx);
trx_resurrect_table_locks(
trx, &trx->rsegs.m_redo, undo);
trx_resurrect_table_locks(trx, undo);
}
/* Ressurrect transactions that were doing updates. */
for (undo = UT_LIST_GET_FIRST(rseg->update_undo_list);
/* Ressurrect other transactions. */
for (undo = UT_LIST_GET_FIRST(rseg->undo_list);
undo != NULL;
undo = UT_LIST_GET_NEXT(undo_list, undo)) {
@ -1049,6 +1028,7 @@ trx_lists_init_at_db_start()
if (trx == NULL) {
trx = trx_allocate_for_background();
trx->start_time = start_time;
ut_d(trx->start_file = __FILE__);
ut_d(trx->start_line = __LINE__);
@ -1058,8 +1038,7 @@ trx_lists_init_at_db_start()
trx_sys_rw_trx_add(trx);
trx_resurrect_table_locks(
trx, &trx->rsegs.m_redo, undo);
trx_resurrect_table_locks(trx, undo);
}
}
@ -1363,7 +1342,7 @@ trx_start_low(
/** Set the serialisation number for a persistent committed transaction.
@param[in,out] trx committed transaction with persistent changes
@param[in,out] rseg rollback segment for update_undo, or NULL */
@param[in,out] rseg rollback segment for undo, or NULL */
static
void
trx_serialise(trx_t* trx, trx_rseg_t* rseg)
@ -1439,37 +1418,49 @@ trx_write_serialisation_history(
}
if (!trx->rsegs.m_redo.rseg) {
ut_ad(!trx->rsegs.m_redo.insert_undo);
ut_ad(!trx->rsegs.m_redo.update_undo);
ut_ad(!trx->rsegs.m_redo.undo);
ut_ad(!trx->rsegs.m_redo.old_insert);
return false;
}
trx_undo_t* insert = trx->rsegs.m_redo.insert_undo;
trx_undo_t* update = trx->rsegs.m_redo.update_undo;
trx_undo_t* undo = trx->rsegs.m_redo.undo;
trx_undo_t*& old_insert = trx->rsegs.m_redo.old_insert;
if (!insert && !update) {
if (!undo && !old_insert) {
return false;
}
ut_ad(!trx->read_only);
trx_rseg_t* update_rseg = update ? trx->rsegs.m_redo.rseg : NULL;
trx_rseg_t* undo_rseg = undo ? undo->rseg : NULL;
ut_ad(!undo || undo->rseg == trx->rsegs.m_redo.rseg);
mutex_enter(&trx->rsegs.m_redo.rseg->mutex);
/* Assign the transaction serialisation number and add any
update_undo log to the purge queue. */
trx_serialise(trx, update_rseg);
undo log to the purge queue. */
trx_serialise(trx, undo_rseg);
/* It is not necessary to acquire trx->undo_mutex here because
only a single OS thread is allowed to commit this transaction. */
if (insert) {
trx_undo_set_state_at_finish(insert, mtr);
if (UNIV_LIKELY_NULL(old_insert)) {
page_t* undo_hdr_page = trx_undo_set_state_at_finish(
old_insert, mtr);
trx_rseg_t* rseg = trx->rsegs.m_redo.rseg;
trx_purge_add_update_undo_to_history(trx, undo_hdr_page, mtr);
UT_LIST_REMOVE(rseg->old_insert_list, old_insert);
if (old_insert->state == TRX_UNDO_CACHED) {
UT_LIST_ADD_FIRST(rseg->undo_cached, old_insert);
MONITOR_INC(MONITOR_NUM_UNDO_SLOT_CACHED);
} else {
ut_ad(old_insert->state == TRX_UNDO_TO_PURGE);
trx_undo_mem_free(old_insert);
}
old_insert = NULL;
}
if (update) {
/* The undo logs and possible delete-marked records
for updates and deletes will be purged later. */
if (undo) {
/* The undo logs will be processed and purged later. */
page_t* undo_hdr_page = trx_undo_set_state_at_finish(
update, mtr);
undo, mtr);
trx_undo_update_cleanup(trx, undo_hdr_page, mtr);
}
@ -1786,7 +1777,7 @@ trx_commit_in_memory(
}
}
ut_ad(!trx->rsegs.m_redo.update_undo);
ut_ad(!trx->rsegs.m_redo.undo);
if (trx_rseg_t* rseg = trx->rsegs.m_redo.rseg) {
mutex_enter(&rseg->mutex);
@ -1794,14 +1785,14 @@ trx_commit_in_memory(
--rseg->trx_ref_count;
mutex_exit(&rseg->mutex);
if (trx_undo_t*& insert = trx->rsegs.m_redo.insert_undo) {
if (trx_undo_t*& insert = trx->rsegs.m_redo.old_insert) {
ut_ad(insert->rseg == rseg);
trx_undo_commit_cleanup(insert, false);
insert = NULL;
}
}
ut_ad(!trx->rsegs.m_redo.insert_undo);
ut_ad(!trx->rsegs.m_redo.old_insert);
if (mtr != NULL) {
if (trx_undo_t*& undo = trx->rsegs.m_noredo.undo) {
@ -2012,7 +2003,7 @@ trx_commit(
DBUG_EXECUTE_IF("ib_trx_commit_crash_before_trx_commit_start",
DBUG_SUICIDE(););
if (trx->has_logged()) {
if (trx->has_logged_or_recovered()) {
mtr = &local_mtr;
mtr_start_sync(mtr);
} else {
@ -2034,9 +2025,9 @@ trx_cleanup_at_db_startup(
{
ut_ad(trx->is_recovered);
ut_ad(!trx->rsegs.m_noredo.undo);
ut_ad(!trx->rsegs.m_redo.update_undo);
ut_ad(!trx->rsegs.m_redo.undo);
if (trx_undo_t*& undo = trx->rsegs.m_redo.insert_undo) {
if (trx_undo_t*& undo = trx->rsegs.m_redo.old_insert) {
ut_ad(undo->rseg == trx->rsegs.m_redo.rseg);
trx_undo_commit_cleanup(undo, false);
undo = NULL;
@ -2666,6 +2657,9 @@ static
lsn_t
trx_prepare_low(trx_t* trx)
{
ut_ad(!trx->rsegs.m_redo.old_insert);
ut_ad(!trx->is_recovered);
mtr_t mtr;
/* It is not necessary to acquire trx->undo_mutex here because
@ -2685,15 +2679,15 @@ trx_prepare_low(trx_t* trx)
mtr.commit();
}
trx_undo_t* insert = trx->rsegs.m_redo.insert_undo;
trx_undo_t* update = trx->rsegs.m_redo.update_undo;
trx_undo_t* undo = trx->rsegs.m_redo.undo;
if (!insert && !update) {
if (!undo) {
/* There were no changes to persistent tables. */
return(0);
}
trx_rseg_t* rseg = trx->rsegs.m_redo.rseg;
ut_ad(undo->rseg == rseg);
mtr.start(true);
@ -2703,17 +2697,7 @@ trx_prepare_low(trx_t* trx)
world, at the serialization point of lsn. */
mutex_enter(&rseg->mutex);
if (insert) {
ut_ad(insert->rseg == rseg);
trx_undo_set_state_at_prepare(trx, insert, false, &mtr);
}
if (update) {
ut_ad(update->rseg == rseg);
trx_undo_set_state_at_prepare(trx, update, false, &mtr);
}
trx_undo_set_state_at_prepare(trx, undo, false, &mtr);
mutex_exit(&rseg->mutex);
/* Make the XA PREPARE durable. */

390
storage/innobase/trx/trx0undo.cc
View File

@ -100,7 +100,6 @@ void
trx_undo_page_init(
/*===============*/
page_t* undo_page, /*!< in: undo log segment page */
ulint type, /*!< in: undo log segment type */
mtr_t* mtr); /*!< in: mtr */
/********************************************************************//**
@ -112,26 +111,11 @@ trx_undo_mem_create(
/*================*/
trx_rseg_t* rseg, /*!< in: rollback segment memory object */
ulint id, /*!< in: slot index within rseg */
ulint type, /*!< in: type of the log: TRX_UNDO_INSERT or
TRX_UNDO_UPDATE */
trx_id_t trx_id, /*!< in: id of the trx for which the undo log
is created */
const XID* xid, /*!< in: X/Open XA transaction identification*/
ulint page_no,/*!< in: undo log header page number */
ulint offset);/*!< in: undo log header byte offset on page */
/***************************************************************//**
Initializes a cached insert undo log header page for new use. NOTE that this
function has its own log record type MLOG_UNDO_HDR_REUSE. You must NOT change
the operation of this function!
@return undo log header byte offset on page */
static
ulint
trx_undo_insert_header_reuse(
/*=========================*/
page_t* undo_page, /*!< in/out: insert undo log segment
header page, x-latched */
trx_id_t trx_id, /*!< in: transaction id */
mtr_t* mtr); /*!< in: mtr */
/**********************************************************************//**
If an update undo log can be discarded immediately, this function frees the
space, resetting the page to the proper state for caching. */
@ -341,12 +325,11 @@ void
trx_undo_page_init_log(
/*===================*/
page_t* undo_page, /*!< in: undo log page */
ulint type, /*!< in: undo log type */
mtr_t* mtr) /*!< in: mtr */
{
mlog_write_initial_log_record(undo_page, MLOG_UNDO_INIT, mtr);
mlog_catenate_ulint_compressed(mtr, type);
mlog_catenate_ulint_compressed(mtr, 0);
}
/***********************************************************//**
@ -360,9 +343,9 @@ trx_undo_parse_page_init(
page_t* page, /*!< in: page or NULL */
mtr_t* mtr) /*!< in: mtr or NULL */
{
ulint type;
type = mach_parse_compressed(&ptr, end_ptr);
if (mach_parse_compressed(&ptr, end_ptr)) {
recv_sys->found_corrupt_log = true;
}
if (ptr == NULL) {
@ -370,7 +353,7 @@ trx_undo_parse_page_init(
}
if (page) {
trx_undo_page_init(page, type, mtr);
trx_undo_page_init(page, mtr);
}
return(const_cast<byte*>(ptr));
@ -383,14 +366,13 @@ void
trx_undo_page_init(
/*===============*/
page_t* undo_page, /*!< in: undo log segment page */
ulint type, /*!< in: undo log segment type */
mtr_t* mtr) /*!< in: mtr */
{
trx_upagef_t* page_hdr;
page_hdr = undo_page + TRX_UNDO_PAGE_HDR;
mach_write_to_2(page_hdr + TRX_UNDO_PAGE_TYPE, type);
*reinterpret_cast<uint16*>(page_hdr + TRX_UNDO_PAGE_TYPE) = 0;
mach_write_to_2(page_hdr + TRX_UNDO_PAGE_START,
TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_HDR_SIZE);
@ -399,7 +381,7 @@ trx_undo_page_init(
fil_page_set_type(undo_page, FIL_PAGE_UNDO_LOG);
trx_undo_page_init_log(undo_page, type, mtr);
trx_undo_page_init_log(undo_page, mtr);
}
/***************************************************************//**
@ -413,8 +395,6 @@ trx_undo_seg_create(
trx_rseg_t* rseg MY_ATTRIBUTE((unused)),/*!< in: rollback segment */
trx_rsegf_t* rseg_hdr,/*!< in: rollback segment header, page
x-latched */
ulint type, /*!< in: type of the segment: TRX_UNDO_INSERT or
TRX_UNDO_UPDATE */
ulint* id, /*!< out: slot index within rseg header */
page_t** undo_page,
/*!< out: segment header page x-latched, NULL
@ -435,9 +415,6 @@ trx_undo_seg_create(
ut_ad(rseg_hdr != NULL);
ut_ad(mutex_own(&(rseg->mutex)));
/* fputs(type == TRX_UNDO_INSERT
? "Creating insert undo log segment\n"
: "Creating update undo log segment\n", stderr); */
slot_no = trx_rsegf_undo_find_free(rseg_hdr, mtr);
if (slot_no == ULINT_UNDEFINED) {
@ -477,7 +454,7 @@ trx_undo_seg_create(
page_hdr = *undo_page + TRX_UNDO_PAGE_HDR;
seg_hdr = *undo_page + TRX_UNDO_SEG_HDR;
trx_undo_page_init(*undo_page, type, mtr);
trx_undo_page_init(*undo_page, mtr);
mlog_write_ulint(page_hdr + TRX_UNDO_PAGE_FREE,
TRX_UNDO_SEG_HDR + TRX_UNDO_SEG_HDR_SIZE,
@ -674,23 +651,7 @@ trx_undo_header_add_space_for_xid(
MLOG_2BYTES, mtr);
}
/**********************************************************************//**
Writes the mtr log entry of an undo log header reuse. */
UNIV_INLINE
void
trx_undo_insert_header_reuse_log(
/*=============================*/
const page_t* undo_page, /*!< in: undo log header page */
trx_id_t trx_id, /*!< in: transaction id */
mtr_t* mtr) /*!< in: mtr */
{
mlog_write_initial_log_record(undo_page, MLOG_UNDO_HDR_REUSE, mtr);
mlog_catenate_ull_compressed(mtr, trx_id);
}
/** Parse the redo log entry of an undo log page header create or reuse.
@param[in] type MLOG_UNDO_HDR_CREATE or MLOG_UNDO_HDR_REUSE
/** Parse the redo log entry of an undo log page header create.
@param[in] ptr redo log record
@param[in] end_ptr end of log buffer
@param[in,out] page page frame or NULL
@ -698,7 +659,6 @@ trx_undo_insert_header_reuse_log(
@return end of log record or NULL */
byte*
trx_undo_parse_page_header(
mlog_id_t type,
const byte* ptr,
const byte* end_ptr,
page_t* page,
@ -707,82 +667,13 @@ trx_undo_parse_page_header(
trx_id_t trx_id = mach_u64_parse_compressed(&ptr, end_ptr);
if (ptr != NULL && page != NULL) {
switch (type) {
case MLOG_UNDO_HDR_CREATE:
trx_undo_header_create(page, trx_id, mtr);
return(const_cast<byte*>(ptr));
case MLOG_UNDO_HDR_REUSE:
trx_undo_insert_header_reuse(page, trx_id, mtr);
return(const_cast<byte*>(ptr));
default:
break;
}
ut_ad(0);
trx_undo_header_create(page, trx_id, mtr);
return(const_cast<byte*>(ptr));
}
return(const_cast<byte*>(ptr));
}
/***************************************************************//**
Initializes a cached insert undo log header page for new use. NOTE that this
function has its own log record type MLOG_UNDO_HDR_REUSE. You must NOT change
the operation of this function!
@return undo log header byte offset on page */
static
ulint
trx_undo_insert_header_reuse(
/*=========================*/
page_t* undo_page, /*!< in/out: insert undo log segment
header page, x-latched */
trx_id_t trx_id, /*!< in: transaction id */
mtr_t* mtr) /*!< in: mtr */
{
trx_upagef_t* page_hdr;
trx_usegf_t* seg_hdr;
trx_ulogf_t* log_hdr;
ulint free;
ulint new_free;
ut_ad(mtr && undo_page);
page_hdr = undo_page + TRX_UNDO_PAGE_HDR;
seg_hdr = undo_page + TRX_UNDO_SEG_HDR;
free = TRX_UNDO_SEG_HDR + TRX_UNDO_SEG_HDR_SIZE;
ut_a(free + TRX_UNDO_LOG_XA_HDR_SIZE < UNIV_PAGE_SIZE - 100);
log_hdr = undo_page + free;
new_free = free + TRX_UNDO_LOG_OLD_HDR_SIZE;
/* Insert undo data is not needed after commit: we may free all
the space on the page */
ut_a(mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
+ TRX_UNDO_PAGE_TYPE)
== TRX_UNDO_INSERT);
mach_write_to_2(page_hdr + TRX_UNDO_PAGE_START, new_free);
mach_write_to_2(page_hdr + TRX_UNDO_PAGE_FREE, new_free);
mach_write_to_2(seg_hdr + TRX_UNDO_STATE, TRX_UNDO_ACTIVE);
log_hdr = undo_page + free;
mach_write_to_8(log_hdr + TRX_UNDO_TRX_ID, trx_id);
mach_write_to_2(log_hdr + TRX_UNDO_LOG_START, new_free);
mach_write_to_1(log_hdr + TRX_UNDO_XID_EXISTS, FALSE);
mach_write_to_1(log_hdr + TRX_UNDO_DICT_TRANS, FALSE);
/* Write the log record MLOG_UNDO_HDR_REUSE */
trx_undo_insert_header_reuse_log(undo_page, trx_id, mtr);
return(free);
}
/**********************************************************************//**
Writes the redo log entry of an update undo log header discard. */
UNIV_INLINE
@ -905,7 +796,7 @@ trx_undo_add_page(trx_t* trx, trx_undo_t* undo, mtr_t* mtr)
buf_block_dbg_add_level(new_block, SYNC_TRX_UNDO_PAGE);
undo->last_page_no = new_block->page.id.page_no();
trx_undo_page_init(new_block->frame, undo->type, mtr);
trx_undo_page_init(new_block->frame, mtr);
flst_add_last(TRX_UNDO_SEG_HDR + TRX_UNDO_PAGE_LIST
+ header_page,
@ -1212,28 +1103,20 @@ trx_undo_mem_create_at_db_start(
mtr_t* mtr) /*!< in: mtr */
{
page_t* undo_page;
trx_upagef_t* page_header;
trx_usegf_t* seg_header;
trx_ulogf_t* undo_header;
trx_undo_t* undo;
ulint type;
ulint state;
trx_id_t trx_id;
ulint offset;
fil_addr_t last_addr;
page_t* last_page;
trx_undo_rec_t* rec;
XID xid;
ibool xid_exists = FALSE;
ut_a(id < TRX_RSEG_N_SLOTS);
undo_page = trx_undo_page_get(page_id_t(rseg->space, page_no), mtr);
page_header = undo_page + TRX_UNDO_PAGE_HDR;
type = mtr_read_ulint(page_header + TRX_UNDO_PAGE_TYPE, MLOG_2BYTES,
mtr);
const ulint type = mach_read_from_2(
TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_TYPE + undo_page);
ut_ad(type == 0 || type == TRX_UNDO_INSERT || type == TRX_UNDO_UPDATE);
seg_header = undo_page + TRX_UNDO_SEG_HDR;
state = mach_read_from_2(seg_header + TRX_UNDO_STATE);
@ -1244,75 +1127,62 @@ trx_undo_mem_create_at_db_start(
trx_id = mach_read_from_8(undo_header + TRX_UNDO_TRX_ID);
xid_exists = mtr_read_ulint(undo_header + TRX_UNDO_XID_EXISTS,
MLOG_1BYTE, mtr);
const bool xid_exists = mtr_read_ulint(
undo_header + TRX_UNDO_XID_EXISTS, MLOG_1BYTE, mtr);
/* Read X/Open XA transaction identification if it exists, or
set it to NULL. */
xid.null();
if (xid_exists == TRUE) {
if (xid_exists) {
trx_undo_read_xid(undo_header, &xid);
}
mutex_enter(&(rseg->mutex));
undo = trx_undo_mem_create(rseg, id, type, trx_id, &xid,
page_no, offset);
undo = trx_undo_mem_create(rseg, id, trx_id, &xid, page_no, offset);
mutex_exit(&(rseg->mutex));
undo->dict_operation = mtr_read_ulint(
undo_header + TRX_UNDO_DICT_TRANS, MLOG_1BYTE, mtr);
undo->table_id = mach_read_from_8(undo_header + TRX_UNDO_TABLE_ID);
undo->state = state;
undo->size = flst_get_len(seg_header + TRX_UNDO_PAGE_LIST);
/* If the log segment is being freed, the page list is inconsistent! */
if (state == TRX_UNDO_TO_FREE) {
goto add_to_list;
}
last_addr = flst_get_last(seg_header + TRX_UNDO_PAGE_LIST, mtr);
if (UNIV_UNLIKELY(state == TRX_UNDO_TO_FREE)) {
/* This is an old-format insert_undo log segment that
is being freed. The page list is inconsistent. */
ut_ad(type == TRX_UNDO_INSERT);
state = TRX_UNDO_TO_PURGE;
} else {
fil_addr_t last_addr = flst_get_last(
seg_header + TRX_UNDO_PAGE_LIST, mtr);
undo->last_page_no = last_addr.page;
undo->top_page_no = last_addr.page;
undo->last_page_no = last_addr.page;
undo->top_page_no = last_addr.page;
last_page = trx_undo_page_get(
page_id_t(rseg->space, undo->last_page_no), mtr);
page_t* last_page = trx_undo_page_get(
page_id_t(rseg->space, undo->last_page_no), mtr);
rec = trx_undo_page_get_last_rec(last_page, page_no, offset);
const trx_undo_rec_t* rec = trx_undo_page_get_last_rec(
last_page, page_no, offset);
if (rec == NULL) {
undo->empty = TRUE;
} else {
undo->empty = FALSE;
undo->top_offset = rec - last_page;
undo->top_undo_no = trx_undo_rec_get_undo_no(rec);
undo->empty = !rec;
if (rec) {
undo->top_offset = rec - last_page;
undo->top_undo_no = trx_undo_rec_get_undo_no(rec);
}
}
add_to_list:
if (type == TRX_UNDO_INSERT) {
if (state != TRX_UNDO_CACHED) {
UT_LIST_ADD_LAST(rseg->insert_undo_list, undo);
} else {
UT_LIST_ADD_LAST(rseg->insert_undo_cached, undo);
undo->state = state;
MONITOR_INC(MONITOR_NUM_UNDO_SLOT_CACHED);
}
if (state != TRX_UNDO_CACHED) {
UT_LIST_ADD_LAST(type == TRX_UNDO_INSERT
? rseg->old_insert_list
: rseg->undo_list, undo);
} else {
ut_ad(type == TRX_UNDO_UPDATE);
if (state != TRX_UNDO_CACHED) {
UT_LIST_ADD_LAST(rseg->update_undo_list, undo);
} else {
UT_LIST_ADD_LAST(rseg->update_undo_cached, undo);
MONITOR_INC(MONITOR_NUM_UNDO_SLOT_CACHED);
}
UT_LIST_ADD_LAST(rseg->undo_cached, undo);
MONITOR_INC(MONITOR_NUM_UNDO_SLOT_CACHED);
}
return(undo);
@ -1383,8 +1253,6 @@ trx_undo_mem_create(
/*================*/
trx_rseg_t* rseg, /*!< in: rollback segment memory object */
ulint id, /*!< in: slot index within rseg */
ulint type, /*!< in: type of the log: TRX_UNDO_INSERT or
TRX_UNDO_UPDATE */
trx_id_t trx_id, /*!< in: id of the trx for which the undo log
is created */
const XID* xid, /*!< in: X/Open transaction identification */
@ -1405,7 +1273,6 @@ trx_undo_mem_create(
}
undo->id = id;
undo->type = type;
undo->state = TRX_UNDO_ACTIVE;
undo->del_marks = FALSE;
undo->trx_id = trx_id;
@ -1479,8 +1346,6 @@ trx_undo_create(
/*============*/
trx_t* trx, /*!< in: transaction */
trx_rseg_t* rseg, /*!< in: rollback segment memory copy */
ulint type, /*!< in: type of the log: TRX_UNDO_INSERT or
TRX_UNDO_UPDATE */
trx_id_t trx_id, /*!< in: id of the trx for which the undo log
is created */
const XID* xid, /*!< in: X/Open transaction identification*/
@ -1506,8 +1371,7 @@ trx_undo_create(
rseg_header = trx_rsegf_get(rseg->space, rseg->page_no, mtr);
err = trx_undo_seg_create(rseg, rseg_header, type, &id,
&undo_page, mtr);
err = trx_undo_seg_create(rseg, rseg_header, &id, &undo_page, mtr);
if (err != DB_SUCCESS) {
/* Did not succeed */
@ -1523,8 +1387,7 @@ trx_undo_create(
trx_undo_header_add_space_for_xid(undo_page, undo_page + offset, mtr);
*undo = trx_undo_mem_create(rseg, id, type, trx_id, xid,
page_no, offset);
*undo = trx_undo_mem_create(rseg, id, trx_id, xid, page_no, offset);
if (*undo == NULL) {
err = DB_OUT_OF_MEMORY;
@ -1544,8 +1407,6 @@ trx_undo_reuse_cached(
/*==================*/
trx_t* trx, /*!< in: transaction */
trx_rseg_t* rseg, /*!< in: rollback segment memory object */
ulint type, /*!< in: type of the log: TRX_UNDO_INSERT or
TRX_UNDO_UPDATE */
trx_id_t trx_id, /*!< in: id of the trx for which the undo log
is used */
const XID* xid, /*!< in: X/Open XA transaction identification */
@ -1557,30 +1418,12 @@ trx_undo_reuse_cached(
ut_ad(mutex_own(&(rseg->mutex)));
if (type == TRX_UNDO_INSERT) {
undo = UT_LIST_GET_FIRST(rseg->insert_undo_cached);
if (undo == NULL) {
return(NULL);
}
UT_LIST_REMOVE(rseg->insert_undo_cached, undo);
MONITOR_DEC(MONITOR_NUM_UNDO_SLOT_CACHED);
} else {
ut_ad(type == TRX_UNDO_UPDATE);
undo = UT_LIST_GET_FIRST(rseg->update_undo_cached);
if (undo == NULL) {
return(NULL);
}
UT_LIST_REMOVE(rseg->update_undo_cached, undo);
MONITOR_DEC(MONITOR_NUM_UNDO_SLOT_CACHED);
undo = UT_LIST_GET_FIRST(rseg->undo_cached);
if (undo == NULL) {
return(NULL);
}
UT_LIST_REMOVE(rseg->undo_cached, undo);
MONITOR_DEC(MONITOR_NUM_UNDO_SLOT_CACHED);
ut_ad(undo->size == 1);
ut_a(undo->id < TRX_RSEG_N_SLOTS);
@ -1588,22 +1431,9 @@ trx_undo_reuse_cached(
undo_page = trx_undo_page_get(
page_id_t(undo->space, undo->hdr_page_no), mtr);
if (type == TRX_UNDO_INSERT) {
offset = trx_undo_insert_header_reuse(undo_page, trx_id, mtr);
trx_undo_header_add_space_for_xid(
undo_page, undo_page + offset, mtr);
} else {
ut_a(mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
+ TRX_UNDO_PAGE_TYPE)
== TRX_UNDO_UPDATE);
offset = trx_undo_header_create(undo_page, trx_id, mtr);
trx_undo_header_add_space_for_xid(
undo_page, undo_page + offset, mtr);
}
offset = trx_undo_header_create(undo_page, trx_id, mtr);
trx_undo_header_add_space_for_xid(undo_page, undo_page + offset, mtr);
trx_undo_mem_init_for_reuse(undo, trx_id, xid, offset);
return(undo);
@ -1652,18 +1482,13 @@ A new undo log is created or a cached undo log reused.
@param[in,out] trx transaction
@param[in] rseg rollback segment
@param[out] undo the undo log
@param[in] type TRX_UNDO_INSERT or TRX_UNDO_UPDATE
@retval DB_SUCCESS on success
@retval DB_TOO_MANY_CONCURRENT_TRXS
@retval DB_OUT_OF_FILE_SPACE
@retval DB_READ_ONLY
@retval DB_OUT_OF_MEMORY */
dberr_t
trx_undo_assign_undo(
trx_t* trx,
trx_rseg_t* rseg,
trx_undo_t** undo,
ulint type)
trx_undo_assign_undo(trx_t* trx, trx_rseg_t* rseg, trx_undo_t** undo)
{
const bool is_temp = rseg == trx->rsegs.m_noredo.rseg;
mtr_t mtr;
@ -1672,40 +1497,31 @@ trx_undo_assign_undo(
ut_ad(mutex_own(&trx->undo_mutex));
ut_ad(rseg == trx->rsegs.m_redo.rseg
|| rseg == trx->rsegs.m_noredo.rseg);
ut_ad(type == TRX_UNDO_INSERT || type == TRX_UNDO_UPDATE);
ut_ad(undo == (is_temp
? &trx->rsegs.m_noredo.undo
: &trx->rsegs.m_redo.undo));
mtr.start(trx);
if (is_temp) {
mtr.set_log_mode(MTR_LOG_NO_REDO);
ut_ad(undo == &trx->rsegs.m_noredo.undo);
} else {
ut_ad(undo == (type == TRX_UNDO_INSERT
? &trx->rsegs.m_redo.insert_undo
: &trx->rsegs.m_redo.update_undo));
}
mutex_enter(&rseg->mutex);
*undo = trx_undo_reuse_cached(trx, rseg, type, trx->id, trx->xid,
&mtr);
*undo = trx_undo_reuse_cached(trx, rseg, trx->id, trx->xid, &mtr);
if (*undo == NULL) {
err = trx_undo_create(trx, rseg, type, trx->id, trx->xid,
err = trx_undo_create(trx, rseg, trx->id, trx->xid,
undo, &mtr);
if (err != DB_SUCCESS) {
goto func_exit;
}
}
if (is_temp) {
UT_LIST_ADD_FIRST(rseg->insert_undo_list, *undo);
} else {
UT_LIST_ADD_FIRST(type == TRX_UNDO_INSERT
? rseg->insert_undo_list
: rseg->update_undo_list, *undo);
if (trx_get_dict_operation(trx) != TRX_DICT_OP_NONE) {
trx_undo_mark_as_dict_operation(trx, *undo, &mtr);
}
UT_LIST_ADD_FIRST(rseg->undo_list, *undo);
if (!is_temp && trx_get_dict_operation(trx) != TRX_DICT_OP_NONE) {
trx_undo_mark_as_dict_operation(trx, *undo, &mtr);
}
func_exit:
@ -1742,10 +1558,6 @@ trx_undo_set_state_at_finish(
< TRX_UNDO_PAGE_REUSE_LIMIT) {
state = TRX_UNDO_CACHED;
} else if (undo->type == TRX_UNDO_INSERT) {
state = TRX_UNDO_TO_FREE;
} else {
state = TRX_UNDO_TO_PURGE;
}
@ -1759,7 +1571,7 @@ trx_undo_set_state_at_finish(
/** Set the state of the undo log segment at a XA PREPARE or XA ROLLBACK.
@param[in,out] trx transaction
@param[in,out] undo insert_undo or update_undo log
@param[in,out] undo undo log
@param[in] rollback false=XA PREPARE, true=XA ROLLBACK
@param[in,out] mtr mini-transaction
@return undo log segment header page, x-latched */
@ -1824,20 +1636,20 @@ trx_undo_update_cleanup(
x-latched */
mtr_t* mtr) /*!< in: mtr */
{
trx_undo_t* undo = trx->rsegs.m_redo.update_undo;
trx_undo_t* undo = trx->rsegs.m_redo.undo;
trx_rseg_t* rseg = undo->rseg;
ut_ad(mutex_own(&rseg->mutex));
trx_purge_add_update_undo_to_history(trx, undo_page, mtr);
UT_LIST_REMOVE(rseg->update_undo_list, undo);
UT_LIST_REMOVE(rseg->undo_list, undo);
trx->rsegs.m_redo.update_undo = NULL;
trx->rsegs.m_redo.undo = NULL;
if (undo->state == TRX_UNDO_CACHED) {
UT_LIST_ADD_FIRST(rseg->update_undo_cached, undo);
UT_LIST_ADD_FIRST(rseg->undo_cached, undo);
MONITOR_INC(MONITOR_NUM_UNDO_SLOT_CACHED);
} else {
@ -1847,7 +1659,7 @@ trx_undo_update_cleanup(
}
}
/** Free an insert or temporary undo log after commit or rollback.
/** Free an old insert or temporary undo log after commit or rollback.
The information is not needed after a commit or rollback, therefore
the data can be discarded.
@param[in,out] undo undo log
@ -1857,20 +1669,22 @@ trx_undo_commit_cleanup(trx_undo_t* undo, bool is_temp)
{
trx_rseg_t* rseg = undo->rseg;
ut_ad(is_temp == !rseg->is_persistent());
ut_ad(!is_temp || 0 == UT_LIST_GET_LEN(rseg->old_insert_list));
mutex_enter(&rseg->mutex);
UT_LIST_REMOVE(rseg->insert_undo_list, undo);
UT_LIST_REMOVE(is_temp ? rseg->undo_list : rseg->old_insert_list,
undo);
if (undo->state == TRX_UNDO_CACHED) {
UT_LIST_ADD_FIRST(rseg->insert_undo_cached, undo);
UT_LIST_ADD_FIRST(rseg->undo_cached, undo);
MONITOR_INC(MONITOR_NUM_UNDO_SLOT_CACHED);
} else {
ut_ad(undo->state == TRX_UNDO_TO_FREE);
ut_ad(undo->state == TRX_UNDO_TO_PURGE);
/* Delete first the undo log segment in the file */
mutex_exit(&rseg->mutex);
trx_undo_seg_free(undo, is_temp);
trx_undo_seg_free(undo, true);
mutex_enter(&rseg->mutex);
ut_ad(rseg->curr_size > undo->size);
@ -1891,8 +1705,8 @@ trx_undo_free_prepared(
{
ut_ad(srv_shutdown_state == SRV_SHUTDOWN_EXIT_THREADS);
if (trx->rsegs.m_redo.update_undo) {
switch (trx->rsegs.m_redo.update_undo->state) {
if (trx_undo_t*& undo = trx->rsegs.m_redo.undo) {
switch (undo->state) {
case TRX_UNDO_PREPARED:
break;
case TRX_UNDO_ACTIVE:
@ -1906,15 +1720,13 @@ trx_undo_free_prepared(
ut_error;
}
UT_LIST_REMOVE(trx->rsegs.m_redo.rseg->update_undo_list,
trx->rsegs.m_redo.update_undo);
trx_undo_mem_free(trx->rsegs.m_redo.update_undo);
trx->rsegs.m_redo.update_undo = NULL;
UT_LIST_REMOVE(trx->rsegs.m_redo.rseg->undo_list, undo);
trx_undo_mem_free(undo);
undo = NULL;
}
if (trx->rsegs.m_redo.insert_undo) {
switch (trx->rsegs.m_redo.insert_undo->state) {
if (trx_undo_t*& undo = trx->rsegs.m_redo.old_insert) {
switch (undo->state) {
case TRX_UNDO_PREPARED:
break;
case TRX_UNDO_ACTIVE:
@ -1928,18 +1740,15 @@ trx_undo_free_prepared(
ut_error;
}
UT_LIST_REMOVE(trx->rsegs.m_redo.rseg->insert_undo_list,
trx->rsegs.m_redo.insert_undo);
trx_undo_mem_free(trx->rsegs.m_redo.insert_undo);
trx->rsegs.m_redo.insert_undo = NULL;
UT_LIST_REMOVE(trx->rsegs.m_redo.rseg->old_insert_list, undo);
trx_undo_mem_free(undo);
undo = NULL;
}
if (trx_undo_t*& undo = trx->rsegs.m_noredo.undo) {
ut_a(undo->state == TRX_UNDO_PREPARED);
UT_LIST_REMOVE(trx->rsegs.m_noredo.rseg->insert_undo_list,
undo);
UT_LIST_REMOVE(trx->rsegs.m_noredo.rseg->undo_list, undo);
trx_undo_mem_free(undo);
undo = NULL;
}
@ -1991,37 +1800,22 @@ trx_undo_truncate_tablespace(
/* Before re-initialization ensure that we free the existing
structure. There can't be any active transactions. */
ut_a(UT_LIST_GET_LEN(rseg->update_undo_list) == 0);
ut_a(UT_LIST_GET_LEN(rseg->insert_undo_list) == 0);
ut_a(UT_LIST_GET_LEN(rseg->undo_list) == 0);
trx_undo_t* next_undo;
for (trx_undo_t* undo =
UT_LIST_GET_FIRST(rseg->update_undo_cached);
undo != NULL;
undo = next_undo) {
next_undo = UT_LIST_GET_NEXT(undo_list, undo);
UT_LIST_REMOVE(rseg->update_undo_cached, undo);
MONITOR_DEC(MONITOR_NUM_UNDO_SLOT_CACHED);
trx_undo_mem_free(undo);
}
for (trx_undo_t* undo =
UT_LIST_GET_FIRST(rseg->insert_undo_cached);
for (trx_undo_t* undo = UT_LIST_GET_FIRST(rseg->undo_cached);
undo != NULL;
undo = next_undo) {
next_undo = UT_LIST_GET_NEXT(undo_list, undo);
UT_LIST_REMOVE(rseg->insert_undo_cached, undo);
UT_LIST_REMOVE(rseg->undo_cached, undo);
MONITOR_DEC(MONITOR_NUM_UNDO_SLOT_CACHED);
trx_undo_mem_free(undo);
}
UT_LIST_INIT(rseg->update_undo_list, &trx_undo_t::undo_list);
UT_LIST_INIT(rseg->update_undo_cached, &trx_undo_t::undo_list);
UT_LIST_INIT(rseg->insert_undo_list, &trx_undo_t::undo_list);
UT_LIST_INIT(rseg->insert_undo_cached, &trx_undo_t::undo_list);
UT_LIST_INIT(rseg->undo_list, &trx_undo_t::undo_list);
UT_LIST_INIT(rseg->undo_cached, &trx_undo_t::undo_list);
rseg->max_size = mtr_read_ulint(
rseg_header + TRX_RSEG_MAX_SIZE, MLOG_4BYTES, &mtr);

Write Preview
Loading…
Cancel
Save