mariadb/sql/wsrep_binlog.cc

401 lines
11 KiB
C++
Raw Normal View History

/* Copyright (C) 2013 Codership Oy <info@codership.com>
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.,
2019-05-11 21:19:05 +02:00
51 Franklin Street, Fifth Floor, Boston, MA 02110-1335 USA. */
#include "mariadb.h"
2019-01-23 12:30:00 +01:00
#include "mysql/service_wsrep.h"
#include "wsrep_binlog.h"
#include "log.h"
10.4 wsrep group commit fixes (#1224) * MDEV-16509 Improve wsrep commit performance with binlog disabled Release commit order critical section early after trx_commit_low() if binlog is not transaction coordinator. In order to avoid two phase commit, binlog_hton is not registered for THD during IO_CACHE population. Implemented a test which verifies that the transactions release commit order early. This optimization will change behavior during recovery as the commit is not two phase when binlog is off. Fixed and recorded wsrep-recover-v25 and wsrep-recover to match the behavior. * MDEV-18730 Ordering for wsrep binlog group commit Previously out of order execution was allowed for wsrep commits. Established proper ordering by populating wait_for_commit for every wsrep THD and making group commit leader to wait for prior commits before proceeding to trx_group_commit_leader(). * MDEV-18730 Added a test case to verify correct commit ordering * MDEV-16509, MDEV-18730 Review fixes Use WSREP_EMULATE_BINLOG() macro to decide if the binlog_hton should be registered. Whitespace/syntax fixes and cleanups. * MDEV-16509 Require binlog for galera_var_innodb_disallow_writes test If the commit to InnoDB is done in one phase, the native InnoDB behavior is that the transaction is committed in memory before it is persisted to disk. This means that the innodb_disallow_writes=ON may not prevent transaction to become visible to other readers before commit is completely over. On the other hand, if the commit is two phase (as it is with binlog), the transaction will be blocked in prepare phase. Fixed the test to use binlog, which enforces two phase commit, which in turn makes commit to block before the changes become visible to other connections. This guarantees that the test produces expected result.
2019-03-15 06:09:13 +01:00
#include "slave.h"
#include "log_event.h"
#include "wsrep_applier.h"
#include "wsrep_mysqld.h"
2019-01-23 12:30:00 +01:00
#include "transaction.h"
extern handlerton *binlog_hton;
/*
Write the contents of a cache to a memory buffer.
This function quite the same as MYSQL_BIN_LOG::write_cache(),
with the exception that here we write in buffer instead of log file.
*/
int wsrep_write_cache_buf(IO_CACHE *cache, uchar **buf, size_t *buf_len)
{
*buf= NULL;
*buf_len= 0;
my_off_t const saved_pos(my_b_tell(cache));
DBUG_ENTER("wsrep_write_cache_buf");
if (reinit_io_cache(cache, READ_CACHE, 0, 0, 0))
{
WSREP_ERROR("failed to initialize io-cache");
DBUG_RETURN(ER_ERROR_ON_WRITE);
}
2019-01-23 12:30:00 +01:00
uint length= my_b_bytes_in_cache(cache);
if (unlikely(0 == length)) length= my_b_fill(cache);
2019-01-23 12:30:00 +01:00
size_t total_length= 0;
if (likely(length > 0)) do
{
total_length += length;
/*
Bail out if buffer grows too large.
A temporary fix to avoid allocating indefinitely large buffer,
not a real limit on a writeset size which includes other things
like header and keys.
*/
if (total_length > wsrep_max_ws_size)
{
WSREP_WARN("transaction size limit (%lu) exceeded: %zu",
wsrep_max_ws_size, total_length);
goto error;
}
uchar* tmp= (uchar *)my_realloc(PSI_INSTRUMENT_ME, *buf, total_length,
MYF(MY_ALLOW_ZERO_PTR));
if (!tmp)
{
WSREP_ERROR("could not (re)allocate buffer: %zu + %u",
*buf_len, length);
goto error;
}
2019-01-23 12:30:00 +01:00
*buf= tmp;
memcpy(*buf + *buf_len, cache->read_pos, length);
2019-01-23 12:30:00 +01:00
*buf_len= total_length;
if (cache->file < 0)
{
cache->read_pos= cache->read_end;
break;
}
2019-01-23 12:30:00 +01:00
} while ((length= my_b_fill(cache)));
if (reinit_io_cache(cache, WRITE_CACHE, saved_pos, 0, 0))
{
WSREP_WARN("failed to initialize io-cache");
goto cleanup;
}
DBUG_RETURN(0);
error:
if (reinit_io_cache(cache, WRITE_CACHE, saved_pos, 0, 0))
{
WSREP_WARN("failed to initialize io-cache");
}
cleanup:
my_free(*buf);
*buf= NULL;
*buf_len= 0;
DBUG_RETURN(ER_ERROR_ON_WRITE);
}
#define STACK_SIZE 4096 /* 4K - for buffer preallocated on the stack:
* many transactions would fit in there
* so there is no need to reach for the heap */
/*
Write the contents of a cache to wsrep provider.
This function quite the same as MYSQL_BIN_LOG::write_cache(),
with the exception that here we write in buffer instead of log file.
This version uses incremental data appending as it reads it from cache.
*/
2019-01-23 12:30:00 +01:00
static int wsrep_write_cache_inc(THD* const thd,
IO_CACHE* const cache,
size_t* const len)
{
2019-01-23 12:30:00 +01:00
DBUG_ENTER("wsrep_write_cache_inc");
my_off_t const saved_pos(my_b_tell(cache));
if (reinit_io_cache(cache, READ_CACHE, thd->wsrep_sr().log_position(), 0, 0))
2019-01-23 12:30:00 +01:00
{
WSREP_ERROR("failed to initialize io-cache");
DBUG_RETURN(1);;
}
2019-01-23 12:30:00 +01:00
int ret= 0;
size_t total_length(0);
2019-01-23 12:30:00 +01:00
uint length(my_b_bytes_in_cache(cache));
if (unlikely(0 == length)) length= my_b_fill(cache);
2019-01-23 12:30:00 +01:00
if (likely(length > 0))
{
do
{
2019-01-23 12:30:00 +01:00
total_length += length;
/* bail out if buffer grows too large
not a real limit on a writeset size which includes other things
like header and keys.
*/
if (unlikely(total_length > wsrep_max_ws_size))
{
WSREP_WARN("transaction size limit (%lu) exceeded: %zu",
wsrep_max_ws_size, total_length);
ret= 1;
goto cleanup;
}
if (thd->wsrep_cs().append_data(wsrep::const_buffer(cache->read_pos, length)))
goto cleanup;
cache->read_pos= cache->read_end;
} while ((cache->file >= 0) && (length= my_b_fill(cache)));
if (ret == 0)
{
assert(total_length + thd->wsrep_sr().log_position() == saved_pos);
}
2019-01-23 12:30:00 +01:00
}
cleanup:
2019-01-23 12:30:00 +01:00
*len= total_length;
if (reinit_io_cache(cache, WRITE_CACHE, saved_pos, 0, 0))
{
WSREP_ERROR("failed to reinitialize io-cache");
}
DBUG_RETURN(ret);
}
/*
Write the contents of a cache to wsrep provider.
This function quite the same as MYSQL_BIN_LOG::write_cache(),
with the exception that here we write in buffer instead of log file.
*/
2019-01-23 12:30:00 +01:00
int wsrep_write_cache(THD* const thd,
IO_CACHE* const cache,
size_t* const len)
{
2019-01-23 12:30:00 +01:00
return wsrep_write_cache_inc(thd, cache, len);
}
void wsrep_dump_rbr_buf(THD *thd, const void* rbr_buf, size_t buf_len)
{
2018-08-02 07:19:57 +02:00
int len= snprintf(NULL, 0, "%s/GRA_%lld_%lld.log",
wsrep_data_home_dir, (longlong) thd->thread_id,
(longlong) wsrep_thd_trx_seqno(thd));
if (len < 0)
{
WSREP_ERROR("snprintf error: %d, skipping dump.", len);
return;
}
/*
len doesn't count the \0 end-of-string. Use len+1 below
to alloc and pass as an argument to snprintf.
*/
char *filename= (char *)malloc(len+1);
2018-08-02 07:19:57 +02:00
int len1= snprintf(filename, len+1, "%s/GRA_%lld_%lld.log",
wsrep_data_home_dir, (longlong) thd->thread_id,
(long long)wsrep_thd_trx_seqno(thd));
if (len > len1)
{
WSREP_ERROR("RBR dump path truncated: %d, skipping dump.", len);
free(filename);
return;
}
FILE *of= fopen(filename, "wb");
if (of)
{
if (fwrite(rbr_buf, buf_len, 1, of) == 0)
WSREP_ERROR("Failed to write buffer of length %llu to '%s'",
(unsigned long long)buf_len, filename);
fclose(of);
}
else
{
WSREP_ERROR("Failed to open file '%s': %d (%s)",
filename, errno, strerror(errno));
}
free(filename);
}
/* Dump replication buffer along with header to a file. */
void wsrep_dump_rbr_buf_with_header(THD *thd, const void *rbr_buf,
size_t buf_len)
{
DBUG_ENTER("wsrep_dump_rbr_buf_with_header");
File file;
IO_CACHE cache;
Log_event_writer writer(&cache, 0);
Format_description_log_event *ev= 0;
longlong thd_trx_seqno= (long long)wsrep_thd_trx_seqno(thd);
2018-08-02 07:19:57 +02:00
int len= snprintf(NULL, 0, "%s/GRA_%lld_%lld_v2.log",
wsrep_data_home_dir, (longlong)thd->thread_id,
thd_trx_seqno);
/*
len doesn't count the \0 end-of-string. Use len+1 below
to alloc and pass as an argument to snprintf.
*/
char *filename;
if (len < 0 || !(filename= (char*)malloc(len+1)))
{
WSREP_ERROR("snprintf error: %d, skipping dump.", len);
DBUG_VOID_RETURN;
}
2018-08-02 07:19:57 +02:00
int len1= snprintf(filename, len+1, "%s/GRA_%lld_%lld_v2.log",
wsrep_data_home_dir, (longlong) thd->thread_id,
thd_trx_seqno);
if (len > len1)
{
WSREP_ERROR("RBR dump path truncated: %d, skipping dump.", len);
free(filename);
DBUG_VOID_RETURN;
}
if ((file= mysql_file_open(key_file_wsrep_gra_log, filename,
O_RDWR | O_CREAT | O_BINARY, MYF(MY_WME))) < 0)
{
WSREP_ERROR("Failed to open file '%s' : %d (%s)",
filename, errno, strerror(errno));
goto cleanup1;
}
if (init_io_cache(&cache, file, 0, WRITE_CACHE, 0, 0, MYF(MY_WME | MY_NABP)))
{
goto cleanup2;
}
if (my_b_safe_write(&cache, BINLOG_MAGIC, BIN_LOG_HEADER_SIZE))
{
goto cleanup2;
}
/*
Instantiate an FDLE object for non-wsrep threads (to be written
to the dump file).
*/
ev= (thd->wsrep_applier) ? wsrep_get_apply_format(thd) :
(new Format_description_log_event(4));
if (writer.write(ev) || my_b_write(&cache, (uchar*)rbr_buf, buf_len) ||
flush_io_cache(&cache))
{
WSREP_ERROR("Failed to write to '%s'.", filename);
goto cleanup2;
}
cleanup2:
end_io_cache(&cache);
cleanup1:
free(filename);
mysql_file_close(file, MYF(MY_WME));
if (!thd->wsrep_applier) delete ev;
DBUG_VOID_RETURN;
}
2019-01-23 12:30:00 +01:00
int wsrep_write_skip_event(THD* thd)
{
DBUG_ENTER("wsrep_write_skip_event");
Ignorable_log_event skip_event(thd);
int ret= mysql_bin_log.write_event(&skip_event);
if (ret)
{
WSREP_WARN("wsrep_write_skip_event: write to binlog failed: %d", ret);
}
if (!ret && (ret= trans_commit_stmt(thd)))
{
WSREP_WARN("wsrep_write_skip_event: statt commit failed");
}
DBUG_RETURN(ret);
}
int wsrep_write_dummy_event_low(THD *thd, const char *msg)
{
::abort();
return 0;
}
int wsrep_write_dummy_event(THD *orig_thd, const char *msg)
{
return 0;
}
10.4 wsrep group commit fixes (#1224) * MDEV-16509 Improve wsrep commit performance with binlog disabled Release commit order critical section early after trx_commit_low() if binlog is not transaction coordinator. In order to avoid two phase commit, binlog_hton is not registered for THD during IO_CACHE population. Implemented a test which verifies that the transactions release commit order early. This optimization will change behavior during recovery as the commit is not two phase when binlog is off. Fixed and recorded wsrep-recover-v25 and wsrep-recover to match the behavior. * MDEV-18730 Ordering for wsrep binlog group commit Previously out of order execution was allowed for wsrep commits. Established proper ordering by populating wait_for_commit for every wsrep THD and making group commit leader to wait for prior commits before proceeding to trx_group_commit_leader(). * MDEV-18730 Added a test case to verify correct commit ordering * MDEV-16509, MDEV-18730 Review fixes Use WSREP_EMULATE_BINLOG() macro to decide if the binlog_hton should be registered. Whitespace/syntax fixes and cleanups. * MDEV-16509 Require binlog for galera_var_innodb_disallow_writes test If the commit to InnoDB is done in one phase, the native InnoDB behavior is that the transaction is committed in memory before it is persisted to disk. This means that the innodb_disallow_writes=ON may not prevent transaction to become visible to other readers before commit is completely over. On the other hand, if the commit is two phase (as it is with binlog), the transaction will be blocked in prepare phase. Fixed the test to use binlog, which enforces two phase commit, which in turn makes commit to block before the changes become visible to other connections. This guarantees that the test produces expected result.
2019-03-15 06:09:13 +01:00
bool wsrep_commit_will_write_binlog(THD *thd)
{
return (!wsrep_emulate_bin_log && /* binlog enabled*/
(wsrep_thd_is_local(thd) || /* local thd*/
(thd->wsrep_applier_service && /* applier and log-slave-updates */
opt_log_slave_updates)));
}
/*
The last THD/commit_for_wait registered for group commit.
*/
static wait_for_commit *commit_order_tail= NULL;
void wsrep_register_for_group_commit(THD *thd)
{
DBUG_ENTER("wsrep_register_for_group_commit");
if (wsrep_emulate_bin_log)
{
/* Binlog is off, no need to maintain group commit queue */
DBUG_VOID_RETURN;
}
DBUG_ASSERT(thd->wsrep_trx().ordered());
10.4 wsrep group commit fixes (#1224) * MDEV-16509 Improve wsrep commit performance with binlog disabled Release commit order critical section early after trx_commit_low() if binlog is not transaction coordinator. In order to avoid two phase commit, binlog_hton is not registered for THD during IO_CACHE population. Implemented a test which verifies that the transactions release commit order early. This optimization will change behavior during recovery as the commit is not two phase when binlog is off. Fixed and recorded wsrep-recover-v25 and wsrep-recover to match the behavior. * MDEV-18730 Ordering for wsrep binlog group commit Previously out of order execution was allowed for wsrep commits. Established proper ordering by populating wait_for_commit for every wsrep THD and making group commit leader to wait for prior commits before proceeding to trx_group_commit_leader(). * MDEV-18730 Added a test case to verify correct commit ordering * MDEV-16509, MDEV-18730 Review fixes Use WSREP_EMULATE_BINLOG() macro to decide if the binlog_hton should be registered. Whitespace/syntax fixes and cleanups. * MDEV-16509 Require binlog for galera_var_innodb_disallow_writes test If the commit to InnoDB is done in one phase, the native InnoDB behavior is that the transaction is committed in memory before it is persisted to disk. This means that the innodb_disallow_writes=ON may not prevent transaction to become visible to other readers before commit is completely over. On the other hand, if the commit is two phase (as it is with binlog), the transaction will be blocked in prepare phase. Fixed the test to use binlog, which enforces two phase commit, which in turn makes commit to block before the changes become visible to other connections. This guarantees that the test produces expected result.
2019-03-15 06:09:13 +01:00
wait_for_commit *wfc= thd->wait_for_commit_ptr= &thd->wsrep_wfc;
mysql_mutex_lock(&LOCK_wsrep_group_commit);
if (commit_order_tail)
{
wfc->register_wait_for_prior_commit(commit_order_tail);
}
commit_order_tail= thd->wait_for_commit_ptr;
mysql_mutex_unlock(&LOCK_wsrep_group_commit);
/*
Now we have queued for group commit. If the commit will go
through TC log_and_order(), the commit ordering is done
by TC group commit. Otherwise the wait for prior
commits to complete is done in ha_commit_one_phase().
*/
DBUG_VOID_RETURN;
}
void wsrep_unregister_from_group_commit(THD *thd)
{
DBUG_ASSERT(thd->wsrep_trx().ordered());
10.4 wsrep group commit fixes (#1224) * MDEV-16509 Improve wsrep commit performance with binlog disabled Release commit order critical section early after trx_commit_low() if binlog is not transaction coordinator. In order to avoid two phase commit, binlog_hton is not registered for THD during IO_CACHE population. Implemented a test which verifies that the transactions release commit order early. This optimization will change behavior during recovery as the commit is not two phase when binlog is off. Fixed and recorded wsrep-recover-v25 and wsrep-recover to match the behavior. * MDEV-18730 Ordering for wsrep binlog group commit Previously out of order execution was allowed for wsrep commits. Established proper ordering by populating wait_for_commit for every wsrep THD and making group commit leader to wait for prior commits before proceeding to trx_group_commit_leader(). * MDEV-18730 Added a test case to verify correct commit ordering * MDEV-16509, MDEV-18730 Review fixes Use WSREP_EMULATE_BINLOG() macro to decide if the binlog_hton should be registered. Whitespace/syntax fixes and cleanups. * MDEV-16509 Require binlog for galera_var_innodb_disallow_writes test If the commit to InnoDB is done in one phase, the native InnoDB behavior is that the transaction is committed in memory before it is persisted to disk. This means that the innodb_disallow_writes=ON may not prevent transaction to become visible to other readers before commit is completely over. On the other hand, if the commit is two phase (as it is with binlog), the transaction will be blocked in prepare phase. Fixed the test to use binlog, which enforces two phase commit, which in turn makes commit to block before the changes become visible to other connections. This guarantees that the test produces expected result.
2019-03-15 06:09:13 +01:00
wait_for_commit *wfc= thd->wait_for_commit_ptr;
if (wfc)
{
mysql_mutex_lock(&LOCK_wsrep_group_commit);
wfc->unregister_wait_for_prior_commit();
thd->wakeup_subsequent_commits(0);
/* The last one queued for group commit has completed commit, it is
safe to set tail to NULL. */
if (wfc == commit_order_tail)
commit_order_tail= NULL;
mysql_mutex_unlock(&LOCK_wsrep_group_commit);
thd->wait_for_commit_ptr= NULL;
}
}