Parallel replication (in 10.0 / "conservative" mode) relies on binlog group
commits to group transactions that can be safely run in parallel on the
slave. The --binlog-commit-wait-count and --binlog-commit-wait-usec options
exist to increase the number of commits per group. But in case of conflicts
between transactions, this can cause unnecessary delay and reduced througput,
especially on a slave where commit order is fixed.
This patch adds a heuristics to reduce this problem. When transaction T1 goes
to commit, it will first wait for N transactions to queue up for a group
commit. However, if we detect that another transaction T2 is waiting for a row
lock held by T1, then we will skip the wait and let T1 commit immediately,
releasing locks and let T2 continue.
On a slave, this avoids the unfortunate situation where T1 is waiting for T2
to join the group commit, but T2 is waiting for T1 to release locks, causing
no work to be done for the duration of the --binlog-commit-wait-usec timeout.
(The heuristic seems reasonable on the master as well, so it is enabled for
all transactions, not just replication transactions).
This bug was seen when parallel replication experienced a deadlock between
transactions T1 and T2, where T2 has reached the commit phase and is waiting
for T1 to commit first. In this case, the deadlock is broken by sending a kill
to T2; that kill error is then later detected and converted to a deadlock
error, which causes T2 to be rolled back and retried.
The problem was that the kill caused ha_commit_trans() to errorneously call
wakeup_subsequent_commits() on T3, signalling it to abort because T2 failed
during commit. This is incorrect, because the error in T2 is only a temporary
error, which will be resolved by normal transaction retry. We should not
signal error to the next transaction until we have executed the code that
handles such temporary errors.
So this patch just removes the calls to wakeup_subsequent_commits() from
ha_commit_trans(). They are incorrect in this case, and they are not needed in
general, as wakeup_subsequent_commits() must in any case be called in
finish_event_group() to wakeup any transactions that may have started to wait
after ha_commit_trans(). And normally, wakeup will in fact have happened
earlier, either from the binlog group commit code, or (in case of no
binlogging) after the fast part of InnoDB/XtraDB group commit.
The symptom of this bug was that replication would break on some transaction
with "Commit failed due to failure of an earlier commit on which this one
depends", but with no such failure of an earlier commit visible anywhere.
The code in binlog group commit around wait_for_commit that controls commit
order, did the wakeup of subsequent commits early, as soon as a following
transaction is put into the group commit queue, but before any such commit has
actually taken place. This causes problems with too early wakeup of
transactions that need to wait for prior to commit, but do not take part in
the binlog group commit for one reason or the other.
This patch solves the problem, by moving the wakeup to happen only after the
binlog group commit is completed.
This requires a new solution to ensure that transactions that arrive later
than the leader are still able to participate in group commit. This patch
introduces a flag wait_for_commit::commit_started. When this is set, a waiter
can queue up itself in the group commit queue.
This way, effectively the wait_for_prior_commit() is skipped only for
transactions that participate in group commit, so that skipping the wait is
safe. Other transactions still wait as needed for correctness.
In SAFE_MUTEX builds, reset the wait_for_commit mutex (destroy and
re-initialise), so that SAFE_MUTEX lock order check does not become
confused when the mutex is re-used for a different purpose.
The reason for the failure was a bug in an include file on debian that causes 'struct stat'
to have different sized depending on the environment.
This patch fixes so that we always include my_global.h or my_config.h before we include any other files.
Other things:
- Removed #include <my_global.h> in some include files; Better to always do this at the top level to have as few
"always-include-this-file-first' files as possible.
- Removed usage of some include files that where already included by my_global.h or by other files.
client/mysql_plugin.c:
Use my_global.h first
client/mysqlslap.c:
Remove duplicated include files
extra/comp_err.c:
Remove duplicated include files
include/m_string.h:
Remove duplicated include files
include/maria.h:
Remove duplicated include files
libmysqld/emb_qcache.cc:
Use my_global.h first
plugin/semisync/semisync.h:
Use my_pthread.h first
sql/datadict.cc:
Use my_global.h first
sql/debug_sync.cc:
Use my_global.h first
sql/derror.cc:
Use my_global.h first
sql/des_key_file.cc:
Use my_global.h first
sql/discover.cc:
Use my_global.h first
sql/event_data_objects.cc:
Use my_global.h first
sql/event_db_repository.cc:
Use my_global.h first
sql/event_parse_data.cc:
Use my_global.h first
sql/event_queue.cc:
Use my_global.h first
sql/event_scheduler.cc:
Use my_global.h first
sql/events.cc:
Use my_global.h first
sql/field.cc:
Use my_global.h first
Remove duplicated include files
sql/field_conv.cc:
Use my_global.h first
sql/filesort.cc:
Use my_global.h first
Remove duplicated include files
sql/gstream.cc:
Use my_global.h first
sql/ha_ndbcluster.cc:
Use my_global.h first
sql/ha_ndbcluster_binlog.cc:
Use my_global.h first
sql/ha_ndbcluster_cond.cc:
Use my_global.h first
sql/ha_partition.cc:
Use my_global.h first
sql/handler.cc:
Use my_global.h first
sql/hash_filo.cc:
Use my_global.h first
sql/hostname.cc:
Use my_global.h first
sql/init.cc:
Use my_global.h first
sql/item.cc:
Use my_global.h first
sql/item_buff.cc:
Use my_global.h first
sql/item_cmpfunc.cc:
Use my_global.h first
sql/item_create.cc:
Use my_global.h first
sql/item_geofunc.cc:
Use my_global.h first
sql/item_inetfunc.cc:
Use my_global.h first
sql/item_row.cc:
Use my_global.h first
sql/item_strfunc.cc:
Use my_global.h first
sql/item_subselect.cc:
Use my_global.h first
sql/item_sum.cc:
Use my_global.h first
sql/item_timefunc.cc:
Use my_global.h first
sql/item_xmlfunc.cc:
Use my_global.h first
sql/key.cc:
Use my_global.h first
sql/lock.cc:
Use my_global.h first
sql/log.cc:
Use my_global.h first
sql/log_event.cc:
Use my_global.h first
sql/log_event_old.cc:
Use my_global.h first
sql/mf_iocache.cc:
Use my_global.h first
sql/mysql_install_db.cc:
Remove duplicated include files
sql/mysqld.cc:
Remove duplicated include files
sql/net_serv.cc:
Remove duplicated include files
sql/opt_range.cc:
Use my_global.h first
sql/opt_subselect.cc:
Use my_global.h first
sql/opt_sum.cc:
Use my_global.h first
sql/parse_file.cc:
Use my_global.h first
sql/partition_info.cc:
Use my_global.h first
sql/procedure.cc:
Use my_global.h first
sql/protocol.cc:
Use my_global.h first
sql/records.cc:
Use my_global.h first
sql/records.h:
Don't include my_global.h
Better to do this at the upper level
sql/repl_failsafe.cc:
Use my_global.h first
sql/rpl_filter.cc:
Use my_global.h first
sql/rpl_gtid.cc:
Use my_global.h first
sql/rpl_handler.cc:
Use my_global.h first
sql/rpl_injector.cc:
Use my_global.h first
sql/rpl_record.cc:
Use my_global.h first
sql/rpl_record_old.cc:
Use my_global.h first
sql/rpl_reporting.cc:
Use my_global.h first
sql/rpl_rli.cc:
Use my_global.h first
sql/rpl_tblmap.cc:
Use my_global.h first
sql/rpl_utility.cc:
Use my_global.h first
sql/set_var.cc:
Added comment
sql/slave.cc:
Use my_global.h first
sql/sp.cc:
Use my_global.h first
sql/sp_cache.cc:
Use my_global.h first
sql/sp_head.cc:
Use my_global.h first
sql/sp_pcontext.cc:
Use my_global.h first
sql/sp_rcontext.cc:
Use my_global.h first
sql/spatial.cc:
Use my_global.h first
sql/sql_acl.cc:
Use my_global.h first
sql/sql_admin.cc:
Use my_global.h first
sql/sql_analyse.cc:
Use my_global.h first
sql/sql_audit.cc:
Use my_global.h first
sql/sql_base.cc:
Use my_global.h first
sql/sql_binlog.cc:
Use my_global.h first
sql/sql_bootstrap.cc:
Use my_global.h first
Use my_global.h first
sql/sql_cache.cc:
Use my_global.h first
sql/sql_class.cc:
Use my_global.h first
sql/sql_client.cc:
Use my_global.h first
sql/sql_connect.cc:
Use my_global.h first
sql/sql_crypt.cc:
Use my_global.h first
sql/sql_cursor.cc:
Use my_global.h first
sql/sql_db.cc:
Use my_global.h first
sql/sql_delete.cc:
Use my_global.h first
sql/sql_derived.cc:
Use my_global.h first
sql/sql_do.cc:
Use my_global.h first
sql/sql_error.cc:
Use my_global.h first
sql/sql_explain.cc:
Use my_global.h first
sql/sql_expression_cache.cc:
Use my_global.h first
sql/sql_handler.cc:
Use my_global.h first
sql/sql_help.cc:
Use my_global.h first
sql/sql_insert.cc:
Use my_global.h first
sql/sql_lex.cc:
Use my_global.h first
sql/sql_load.cc:
Use my_global.h first
sql/sql_locale.cc:
Use my_global.h first
sql/sql_manager.cc:
Use my_global.h first
sql/sql_parse.cc:
Use my_global.h first
sql/sql_partition.cc:
Use my_global.h first
sql/sql_plugin.cc:
Added comment
sql/sql_prepare.cc:
Use my_global.h first
sql/sql_priv.h:
Added error if we use this before including my_global.h
This check is here becasue so many files includes sql_priv.h first.
sql/sql_profile.cc:
Use my_global.h first
sql/sql_reload.cc:
Use my_global.h first
sql/sql_rename.cc:
Use my_global.h first
sql/sql_repl.cc:
Use my_global.h first
sql/sql_select.cc:
Use my_global.h first
sql/sql_servers.cc:
Use my_global.h first
sql/sql_show.cc:
Added comment
sql/sql_signal.cc:
Use my_global.h first
sql/sql_statistics.cc:
Use my_global.h first
sql/sql_table.cc:
Use my_global.h first
sql/sql_tablespace.cc:
Use my_global.h first
sql/sql_test.cc:
Use my_global.h first
sql/sql_time.cc:
Use my_global.h first
sql/sql_trigger.cc:
Use my_global.h first
sql/sql_udf.cc:
Use my_global.h first
sql/sql_union.cc:
Use my_global.h first
sql/sql_update.cc:
Use my_global.h first
sql/sql_view.cc:
Use my_global.h first
sql/sys_vars.cc:
Added comment
sql/table.cc:
Use my_global.h first
sql/thr_malloc.cc:
Use my_global.h first
sql/transaction.cc:
Use my_global.h first
sql/uniques.cc:
Use my_global.h first
sql/unireg.cc:
Use my_global.h first
sql/unireg.h:
Removed inclusion of my_global.h
storage/archive/ha_archive.cc:
Added comment
storage/blackhole/ha_blackhole.cc:
Use my_global.h first
storage/csv/ha_tina.cc:
Use my_global.h first
storage/csv/transparent_file.cc:
Use my_global.h first
storage/federated/ha_federated.cc:
Use my_global.h first
storage/federatedx/federatedx_io.cc:
Use my_global.h first
storage/federatedx/federatedx_io_mysql.cc:
Use my_global.h first
storage/federatedx/federatedx_io_null.cc:
Use my_global.h first
storage/federatedx/federatedx_txn.cc:
Use my_global.h first
storage/heap/ha_heap.cc:
Use my_global.h first
storage/innobase/handler/handler0alter.cc:
Use my_global.h first
storage/maria/ha_maria.cc:
Use my_global.h first
storage/maria/unittest/ma_maria_log_cleanup.c:
Remove duplicated include files
storage/maria/unittest/test_file.c:
Added comment
storage/myisam/ha_myisam.cc:
Move sql_plugin.h first as this includes my_global.h
storage/myisammrg/ha_myisammrg.cc:
Use my_global.h first
storage/oqgraph/oqgraph_thunk.cc:
Use my_config.h and my_global.h first
One could not include my_global.h before oqgraph_thunk.h (don't know why)
storage/spider/ha_spider.cc:
Use my_global.h first
storage/spider/hs_client/config.cpp:
Use my_global.h first
storage/spider/hs_client/escape.cpp:
Use my_global.h first
storage/spider/hs_client/fatal.cpp:
Use my_global.h first
storage/spider/hs_client/hstcpcli.cpp:
Use my_global.h first
storage/spider/hs_client/socket.cpp:
Use my_global.h first
storage/spider/hs_client/string_util.cpp:
Use my_global.h first
storage/spider/spd_conn.cc:
Use my_global.h first
storage/spider/spd_copy_tables.cc:
Use my_global.h first
storage/spider/spd_db_conn.cc:
Use my_global.h first
storage/spider/spd_db_handlersocket.cc:
Use my_global.h first
storage/spider/spd_db_mysql.cc:
Use my_global.h first
storage/spider/spd_db_oracle.cc:
Use my_global.h first
storage/spider/spd_direct_sql.cc:
Use my_global.h first
storage/spider/spd_i_s.cc:
Use my_global.h first
storage/spider/spd_malloc.cc:
Use my_global.h first
storage/spider/spd_param.cc:
Use my_global.h first
storage/spider/spd_ping_table.cc:
Use my_global.h first
storage/spider/spd_sys_table.cc:
Use my_global.h first
storage/spider/spd_table.cc:
Use my_global.h first
storage/spider/spd_trx.cc:
Use my_global.h first
storage/xtradb/handler/handler0alter.cc:
Use my_global.h first
storage/xtradb/handler/i_s.cc:
Use my_global.h first
We should assume that the store engine will report the first duplicate key for this case.
Old code of suppression of unsafe logging error with LIMIT didn't work, because of wrong usage of my_interval_timer().
Suppress unsafe logging errors to the error log if we get too many unsafe logging errors in a short time.
This is to not overflow the error log with meaningless errors.
- Each error code is suppressed and counted separately.
- We do a 5 minute suppression of new errors if we get more than 10 errors in that time.
Only print unsafe logging errors if log_warnings > 1.
mysql-test/suite/binlog/r/binlog_stm_unsafe_warning.result:
Update test results as INSERT ... ON DUPLICATE KEY UPDATE doesn't get logged anymore
mysql-test/suite/binlog/r/binlog_unsafe.result:
Update test results as INSERT ... ON DUPLICATE KEY UPDATE doesn't get logged anymore
mysql-test/suite/engines/README:
Fixed typos
mysql-test/suite/rpl/r/rpl_known_bugs_detection.result:
Update test results as INSERT ... ON DUPLICATE KEY UPDATE doesn't get logged anymore
sql/sql_base.cc:
Don't log warning if there are two unique keys used with INSERT .. ON DUPLICATE KEY UPDATE.
We should assume that the store engine will report the first duplicate key for this case.
sql/sql_class.cc:
Suppress error in binary log if we get too many unsafe logging errors in a short time.
Only print unsafe logging errors if log_warnings > 1
MDEV-6560 Assertion `! is_set() ' failed in Diagnostics_area::set_ok_status on killing CREATE OR REPLACE
MDEV-6525 Assertion `table->pos_in_locked _tables == __null || table->pos_in_locked_tables->table = table' failed in mark_used_tables_as_free_for_reuse, locking problems and binlogging problems on CREATE OR REPLACE under lock.
mysql-test/r/create_or_replace.result:
Added test for MDEV-6560
mysql-test/t/create_or_replace.test:
Added test for MDEV-6560
mysql-test/valgrind.supp:
Added suppression for OpenSuse 12.3
sql/sql_base.cc:
More DBUG
sql/sql_class.cc:
Changed that thd_sqlcom_can_generate_row_events() does not report that CREATE OR REPLACE is generating row events.
This is safe as this function is only used by InnoDB/XtraDB to check if a query is generating row events as part of another transaction. As CREATE is always run as it's own transaction, this isn't a problem.
This fixed MDEV-6525.
sql/sql_table.cc:
Remember if reopen_tables() generates an error (which can only happen in case of KILL).
This fixed MDEV-6560
The bug was that in some cases, if a replicated transaction was rolled back
due to deadlock, during the subsequent retry of that transaction, the
gtid_slave_pos would _not_ be updated with the new GTID, leaving the GTID
position of the slave incorrect.
Fix this by ensuring during the retry that we clear the flag that marks that
the GTID has already been recorded in gtid_slave_pos, so that the update of
gtid_slave_pos will be done again during the retry.
In the original bug, the symptom was an assertion due to OPTION_GTID_BEGIN not
being cleared during the retry of the transaction. The reason was some code in
handling of a COMMIT query event, which would not clear the flag when not
recording a GTID in gtid_slave_pos. This commit also fixes that code to always
clear the OPTION_GTID_BEGIN flag for clarity, though it is actually not
possible for OPTION_GTID_BEGIN to become set unless a GTID is pending for
update (after fixing the bug described above).
~40% bugfixed(*) applied
~40$ bugfixed reverted (incorrect or we're not buggy)
~20% bugfixed applied, despite us being not buggy
(*) only changes in the server code, e.g. not cmakefiles
The bug was that my_real_read() called net_before_header_psi() multiple times for long packets.
Fixed by adding a flag when we are reading a header.
Did also some cleanups to interface of my_net_read() to avoid unnecessary calls if performance schema is not used.
- Added my_net_read_packet() as a replacement for my_net_read(). my_net_read() is still in the client library for old clients.
- Removed THD->m_server_idle (not needed anymore as this is now given as argument to my_net_read_packet()
- Added my_net_read_packet(), which is a new version of my_net_read() with a new parameter if we are doing a read for a new command from the server.
- Added tests for compressed protocol and big packets
include/mysql.h.pp:
Added my_net_read_packet() as a replacement for my_net_read()
include/mysql_com.h:
Added my_net_read_packet() as a replacement for my_net_read()
mysql-test/r/mysql_client_test_comp.result:
New test
mysql-test/t/mysql_client_test-master.opt:
Added max_allowed_packet to be able to test big packets and packet size overflows.
mysql-test/t/mysql_client_test_comp-master.opt:
New test
mysql-test/t/mysql_client_test_nonblock-master.opt:
Added max_allowed_packet to be able to test big packets and packet size overflows.
sql-common/client.c:
Use my_net_read_packet()
sql/mf_iocache.cc:
Use my_net_read_packet()
sql/mysqld.cc:
Removed THD->m_server_idle (not needed anymore as this is now given as argument to my_net_read_packet()
sql/net_serv.cc:
Added argument to my_real_read() to indicte if we are reading the first block of the next statement and should call performance schema.
Added 'compatibilty function' my_net_read().
Added my_net_read_packet(), which is a new version of my_net_read() with a new parameter if we are doing a read for a new command from the server.
sql/sql_class.cc:
Removed m_server_idle (not needed anymore)
sql/sql_class.h:
Removed m_server_idle (not needed anymore)
sql/sql_parse.cc:
Removed m_server_idle (not needed anymore)
tests/mysql_client_test.c:
Added tests for compressed protocol and big packets
After-review changes.
For this patch in 10.0, we do not introduce a new public storage engine API,
we just fix the InnoDB/XtraDB issues. In 10.1, we will make a better public
API that can be used for all storage engines (MDEV-6429).
Eliminate the background thread that did deadlock kills asynchroneously.
Instead, we ensure that the InnoDB/XtraDB code can handle doing the kill from
inside the deadlock detection code (when thd_report_wait_for() needs to kill a
later thread to resolve a deadlock).
(We preserve the part of the original patch that introduces dedicated mutex
and condition for the slave init thread, to remove the abuse of
LOCK_thread_count for start/stop synchronisation of the slave init thread).
replication causing replication to fail.
Remove the temporary fix for MDEV-5914, which used READ COMMITTED for parallel
replication worker threads. Replace it with a better, more selective solution.
The issue is with certain edge cases of InnoDB gap locks, for example between
INSERT and ranged DELETE. It is possible for the gap lock set by the DELETE to
block the INSERT, if the DELETE runs first, while the record lock set by
INSERT does not block the DELETE, if the INSERT runs first. This can cause a
conflict between the two in parallel replication on the slave even though they
ran without conflicts on the master.
With this patch, InnoDB will ask the server layer about the two involved
transactions before blocking on a gap lock. If the server layer tells InnoDB
that the transactions are already fixed wrt. commit order, as they are in
parallel replication, InnoDB will ignore the gap lock and allow the two
transactions to proceed in parallel, avoiding the conflict.
Improve the fix for MDEV-6020. When InnoDB itself detects a deadlock, it now
asks the server layer for any preferences about which transaction to roll
back. In case of parallel replication with two transactions T1 and T2 fixed to
commit T1 before T2, the server layer will ask InnoDB to roll back T2 as the
deadlock victim, not T1. This helps in some cases to avoid excessive deadlock
rollback, as T2 will in any case need to wait for T1 to complete before it can
itself commit.
Also some misc. fixes found during development and testing:
- Remove thd_rpl_is_parallel(), it is not used or needed.
- Use KILL_CONNECTION instead of KILL_QUERY when a parallel replication
worker thread is killed to resolve a deadlock with fixed commit
ordering. There are some cases, eg. in sql/sql_parse.cc, where a KILL_QUERY
can be ignored if the query otherwise completed successfully, and this
could cause the deadlock kill to be lost, so that the deadlock was not
correctly resolved.
- Fix random test failure due to missing wait_for_binlog_checkpoint.inc.
- Make sure that deadlock or other temporary errors during parallel
replication are not printed to the the error log; there were some places
around the replication code with extra error logging. These conditions can
occur occasionally and are handled automatically without breaking
replication, so they should not pollute the error log.
- Fix handling of rgi->gtid_sub_id. We need to be able to access this also at
the end of a transaction, to be able to detect and resolve deadlocks due to
commit ordering. But this value was also used as a flag to mark whether
record_gtid() had been called, by being set to zero, losing the value. Now,
introduce a separate flag rgi->gtid_pending, so rgi->gtid_sub_id remains
valid for the entire duration of the transaction.
- Fix one place where the code to handle ignored errors called reset_killed()
unconditionally, even if no error was caught that should be ignored. This
could cause loss of a deadlock kill signal, breaking deadlock detection and
resolution.
- Fix a couple of missing mysql_reset_thd_for_next_command(). This could
cause a prior error condition to remain for the next event executed,
causing assertions about errors already being set and possibly giving
incorrect error handling for following event executions.
- Fix code that cleared thd->rgi_slave in the parallel replication worker
threads after each event execution; this caused the deadlock detection and
handling code to not be able to correctly process the associated
transactions as belonging to replication worker threads.
- Remove useless error code in slave_background_kill_request().
- Fix bug where wfc->wakeup_error was not cleared at
wait_for_commit::unregister_wait_for_prior_commit(). This could cause the
error condition to wrongly propagate to a later wait_for_prior_commit(),
causing spurious ER_PRIOR_COMMIT_FAILED errors.
- Do not put the binlog background thread into the processlist. It causes
too many result differences in mtr, but also it probably is not useful
for users to pollute the process list with a system thread that does not
really perform any user-visible tasks...
replication causing replication to fail.
In parallel replication, we run transactions from the master in parallel, but
force them to commit in the same order they did on the master. If we force T1
to commit before T2, but T2 holds eg. a row lock that is needed by T1, we get
a deadlock when T2 waits until T1 has committed.
Usually, we do not run T1 and T2 in parallel if there is a chance that they
can have conflicting locks like this, but there are certain edge cases where
it can occasionally happen (eg. MDEV-5914, MDEV-5941, MDEV-6020). The bug was
that this would cause replication to hang, eventually getting a lock timeout
and causing the slave to stop with error.
With this patch, InnoDB will report back to the upper layer whenever a
transactions T1 is about to do a lock wait on T2. If T1 and T2 are parallel
replication transactions, and T2 needs to commit later than T1, we can thus
detect the deadlock; we then kill T2, setting a flag that causes it to catch
the kill and convert it to a deadlock error; this error will then cause T2 to
roll back and release its locks (so that T1 can commit), and later T2 will be
re-tried and eventually also committed.
The kill happens asynchroneously in a slave background thread; this is
necessary, as the reporting from InnoDB about lock waits happen deep inside
the locking code, at a point where it is not possible to directly call
THD::awake() due to mutexes held.
Deadlock is assumed to be (very) rarely occuring, so this patch tries to
minimise the performance impact on the normal case where no deadlocks occur,
rather than optimise the handling of the occasional deadlock.
Also fix transaction retry due to deadlock when it happens after a transaction
already signalled to later transactions that it started to commit. In this
case we need to undo this signalling (and later redo it when we commit again
during retry), so following transactions will not start too early.
Also add a missing thd->send_kill_message() that got triggered during testing
(this corrects an incorrect fix for MySQL Bug#58933).
SHOW PROCESSLIST, SHOW BINLOGS
Problem: A deadlock was occurring when 4 threads were
involved in acquiring locks in the following way
Thread 1: Dump thread ( Slave is reconnecting, so on
Master, a new dump thread is trying kill
zombie dump threads. It acquired thread's
LOCK_thd_data and it is about to acquire
mysys_var->current_mutex ( which LOCK_log)
Thread 2: Application thread is executing show binlogs and
acquired LOCK_log and it is about to acquire
LOCK_index.
Thread 3: Application thread is executing Purge binary logs
and acquired LOCK_index and it is about to
acquire LOCK_thread_count.
Thread 4: Application thread is executing show processlist
and acquired LOCK_thread_count and it is
about to acquire zombie dump thread's
LOCK_thd_data.
Deadlock Cycle:
Thread 1 -> Thread 2 -> Thread 3-> Thread 4 ->Thread 1
The same above deadlock was observed even when thread 4 is
executing 'SELECT * FROM information_schema.processlist' command and
acquired LOCK_thread_count and it is about to acquire zombie
dump thread's LOCK_thd_data.
Analysis:
There are four locks involved in the deadlock. LOCK_log,
LOCK_thread_count, LOCK_index and LOCK_thd_data.
LOCK_log, LOCK_thread_count, LOCK_index are global mutexes
where as LOCK_thd_data is local to a thread.
We can divide these four locks in two groups.
Group 1 consists of LOCK_log and LOCK_index and the order
should be LOCK_log followed by LOCK_index.
Group 2 consists of other two mutexes
LOCK_thread_count, LOCK_thd_data and the order should
be LOCK_thread_count followed by LOCK_thd_data.
Unfortunately, there is no specific predefined lock order defined
to follow in the MySQL system when it comes to locks across these
two groups. In the above problematic example,
there is no problem in the way we are acquiring the locks
if you see each thread individually.
But If you combine all 4 threads, they end up in a deadlock.
Fix:
Since everything seems to be fine in the way threads are taking locks,
In this patch We are changing the duration of the locks in Thread 4
to break the deadlock. i.e., before the patch, Thread 4
('show processlist' command) mysqld_list_processes()
function acquires LOCK_thread_count for the complete duration
of the function and it also acquires/releases
each thread's LOCK_thd_data.
LOCK_thread_count is used to protect addition and
deletion of threads in global threads list. While show
process list is looping through all the existing threads,
it will be a problem if a thread is exited but there is no problem
if a new thread is added to the system. Hence a new mutex is
introduced "LOCK_thd_remove" which will protect deletion
of a thread from global threads list. All threads which are
getting exited should acquire LOCK_thd_remove
followed by LOCK_thread_count. (It should take LOCK_thread_count
also because other places of the code still thinks that exit thread
is protected with LOCK_thread_count. In this fix, we are changing
only 'show process list' query logic )
(Eg: unlink_thd logic will be protected with
LOCK_thd_remove).
Logic of mysqld_list_processes(or file_schema_processlist)
will now be protected with 'LOCK_thd_remove' instead of
'LOCK_thread_count'.
Now the new locking order after this patch is:
LOCK_thd_remove -> LOCK_thd_data -> LOCK_log ->
LOCK_index -> LOCK_thread_count
Due to how gap locks work, two transactions could group commit together on the
master, but get lock conflicts and then deadlock due to different thread
scheduling order on slave.
For now, remove these deadlocks by running the parallel slave in READ
COMMITTED mode. And let InnoDB/XtraDB allow statement-based binlogging for the
parallel slave in READ COMMITTED.
We are also investigating a different solution long-term, which is based on
relaxing the gap locks only between the transactions running in parallel for
one slave, but not against possibly external transactions.
Reason for the bug was an optimization for higher connect speed where we moved when global status was updated,
but forgot to update states when slave thread dies.
Fixed by adding thd->add_status_to_global() before deleting slave thread's thd.
mysys/my_delete.c:
Added missing newline
sql/mysqld.cc:
Use add_status_to_global()
sql/slave.cc:
Added missing add_status_to_global()
sql/sql_class.cc:
Use add_status_to_global()
sql/sql_class.h:
Simplify adding local status to global by adding add_status_to_global()
The fill_schema_table() function used to call get_table_share() for a table name in WHERE
then clear the error list. That way plugins receive the superfluous error notification if it
happens in it. Also the problem was that error handler didn't prevent the suppressed
error message from logging anyway as the logging happens in THD::raise_condition
before the handler call.
Trigger_error_handler is remade into Warnings_only_error_handler, so it stores the error
message in all cases in the thd->stmt_da.
Then later the stored error is raised.
Clean up and improve the parallel implementation code, mainly related to
scheduling of work to threads and handling of stop and errors.
Fix a lot of bugs in various corner cases that could lead to crashes or
corruption.
Fix that a single replication domain could easily grab all worker threads and
stall all other domains; now a configuration variable
--slave-domain-parallel-threads allows to limit the number of
workers.
Allow next event group to start as soon as previous group begins the commit
phase (as opposed to when it ends it); this allows multiple event groups on
the slave to participate in group commit, even when no other opportunities for
parallelism are available.
Various fixes:
- Fix some races in the rpl.rpl_parallel test case.
- Fix an old incorrect assertion in Log_event iocache read.
- Fix repeated malloc/free of wait_for_commit and rpl_group_info objects.
- Simplify wait_for_commit wakeup logic.
- Fix one case in queue_for_group_commit() where killing one thread would
fail to correctly signal the error to the next, causing loss of the
transaction after slave restart.
- Fix leaking of pthreads (and their allocated stack) due to missing
PTHREAD_CREATE_DETACHED attribute.
- Fix how one batch of group-committed transactions wait for the previous
batch before starting to execute themselves. The old code had a very
complex scheduling where the first transaction was handled differently,
with subtle bugs in corner cases. Now each event group is always scheduled
for a new worker (in a round-robin fashion amongst available workers).
Keep a count of how many transactions have started to commit, and wait for
that counter to reach the appropriate value.
- Fix slave stop to wait for all workers to actually complete processing;
before, the wait was for update of last_committed_sub_id, which happens a
bit earlier, and could leave worker threads potentially accessing bits of
the replication state that is no longer valid after slave stop.
- Fix a couple of places where the test suite would kill a thread waiting
inside enter_cond() in connection with debug_sync; debug_sync + kill can
crash in rare cases due to a race with mysys_var_current_mutex in this
case.
- Fix some corner cases where we had enter_cond() but no exit_cond().
- Fix that we could get failure in wait_for_prior_commit() but forget to flag
the error with my_error().
- Fix slave stop (both for normal stop and stop due to error). Now, at stop
we pick a specific safe point (in terms of event groups executed) and make
sure that all event groups before that point are executed to completion,
and that no event group after start executing; this ensures a safe place to
restart replication, even for non-transactional stuff/DDL. In error stop,
make sure that all prior event groups are allowed to execute to completion,
and that any later event groups that have started are rolled back, if
possible. The old code could leave eg. T1 and T3 committed but T2 not, or
it could even leave half a transaction not rolled back in some random
worker, which would cause big problems when that worker was later reused
after slave restart.
- Fix the accounting of amount of events queued for one worker. Before, the
amount was reduced immediately as soon as the events were dequeued (which
happens all at once); this allowed twice the amount of events to be queued
in memory for each single worker, which is not what users would expect.
- Fix that an error set during execution of one event was sometimes not
cleared before executing the next, causing problems with the error
reporting.
- Fix incorrect handling of thd->killed in worker threads.
This is port of fix for MySQL BUG#17647863.
revno: 5572
revision-id: jon.hauglid@oracle.com-20131030232243-b0pw98oy72uka2sj
committer: Jon Olav Hauglid <jon.hauglid@oracle.com>
timestamp: Thu 2013-10-31 00:22:43 +0100
message:
Bug#17647863: MYSQL DOES NOT COMPILE ON OSX 10.9 GM
Rename test() macro to MY_TEST() to avoid conflict with libc++.
MDEV-4984: Implement MASTER_GTID_WAIT() and @@LAST_GTID.
MDEV-4726: Race in mysql-test/suite/rpl/t/rpl_gtid_stop_start.test
MDEV-5636: Deadlock in RESET MASTER
MASTER_GTID_WAIT() is similar to MASTER_POS_WAIT(), but works with a
GTID position rather than an old-style filename/offset.
@@LAST_GTID gives the GTID assigned to the last transaction written
into the binlog.
Together, the two can be used by applications to obtain the GTID of
an update on the master, and then do a MASTER_GTID_WAIT() for that
position on any read slave where it is important to get results that
are caught up with the master at least to the point of the update.
The implementation of MASTER_GTID_WAIT() is implemented in a way
that tries to minimise the performance impact on the SQL threads,
even in the presense of many waiters on single GTID positions (as
from @@LAST_GTID).
- CREATE TABLE is by default executed on the slave as CREATE OR REPLACE
- DROP TABLE is by default executed on the slave as DROP TABLE IF NOT EXISTS
This means that a slave will by default continue even if we try to create
a table that existed on the slave (the table will be deleted and re-created) or
if we try to drop a table that didn't exist on the slave.
This should be safe as instead of having the slave stop because of an inconsistency between
master and slave, it will fix the inconsistency.
Those that would prefer to get a stopped slave instead for the above cases can set slave_ddl_exec_mode to STRICT.
- Ensure that a CREATE OR REPLACE TABLE which dropped a table is replicated
- DROP TABLE that generated an error on master is handled as an identical DROP TABLE on the slave (IF NOT EXISTS is not added in this case)
- Added slave_ddl_exec_mode variable to decide how DDL's are replicated
New logic for handling BEGIN GTID ... COMMIT from the binary log:
- When we find a BEGIN GTID, we start a transaction and set OPTION_GTID_BEGIN
- When we find COMMIT, we reset OPTION_GTID_BEGIN and execute the normal COMMIT code.
- While OPTION_GTID_BEGIN is set:
- We don't generate implict commits before or after statements
- All tables are regarded as transactional tables in the binary log (to ensure things are executed exactly as on the master)
- We reset OPTION_GTID_BEGIN also on rollback
This will help ensuring that we don't get any sporadic commits (and thus new GTID's) on the slave and will help keep the GTID's between master and slave in sync.
mysql-test/extra/rpl_tests/rpl_log.test:
Added testing of mode slave_ddl_exec_mode=STRICT
mysql-test/r/mysqld--help.result:
New help messages
mysql-test/suite/rpl/r/create_or_replace_mix.result:
Testing of CREATE OR REPLACE TABLE with replication
mysql-test/suite/rpl/r/create_or_replace_row.result:
Testing of CREATE OR REPLACE TABLE with replication
mysql-test/suite/rpl/r/create_or_replace_statement.result:
Testing replication of create or replace
mysql-test/suite/rpl/r/rpl_gtid_startpos.result:
Test must be run in slave_ddl_exec_mode=STRICT as part of the test depends on that DROP TABLE should fail on slave.
mysql-test/suite/rpl/r/rpl_row_log.result:
Updated result
mysql-test/suite/rpl/r/rpl_row_log_innodb.result:
Updated result
mysql-test/suite/rpl/r/rpl_row_show_relaylog_events.result:
Updated result
mysql-test/suite/rpl/r/rpl_stm_log.result:
Updated result
mysql-test/suite/rpl/r/rpl_stm_mix_show_relaylog_events.result:
Updated result
mysql-test/suite/rpl/r/rpl_temp_table_mix_row.result:
Updated result
mysql-test/suite/rpl/t/create_or_replace.inc:
Testing of CREATE OR REPLACE TABLE with replication
mysql-test/suite/rpl/t/create_or_replace_mix.cnf:
Testing of CREATE OR REPLACE TABLE with replication
mysql-test/suite/rpl/t/create_or_replace_mix.test:
Testing of CREATE OR REPLACE TABLE with replication
mysql-test/suite/rpl/t/create_or_replace_row.cnf:
Testing of CREATE OR REPLACE TABLE with replication
mysql-test/suite/rpl/t/create_or_replace_row.test:
Testing of CREATE OR REPLACE TABLE with replication
mysql-test/suite/rpl/t/create_or_replace_statement.cnf:
Testing of CREATE OR REPLACE TABLE with replication
mysql-test/suite/rpl/t/create_or_replace_statement.test:
Testing of CREATE OR REPLACE TABLE with replication
mysql-test/suite/rpl/t/rpl_gtid_startpos.test:
Test must be run in slave_ddl_exec_mode=STRICT as part of the test depends on that DROP TABLE should fail on slave.
mysql-test/suite/rpl/t/rpl_stm_log.test:
Removed some lines
mysql-test/suite/sys_vars/r/slave_ddl_exec_mode_basic.result:
Testing of slave_ddl_exec_mode
mysql-test/suite/sys_vars/t/slave_ddl_exec_mode_basic.test:
Testing of slave_ddl_exec_mode
sql/handler.cc:
Regard all tables as transactional in commit if OPTION_GTID_BEGIN is set.
This is to ensure that statments are not commited too early if non transactional tables are used.
sql/log.cc:
Regard all tables as transactional in commit if OPTION_GTID_BEGIN is set.
Also treat 'direct' log events as transactional (to get them logged as they where on the master)
sql/log_event.cc:
Ensure that the new error from DROP TABLE when trying to drop a view is treated same as the old one.
Store error code that slave expects in THD.
Set OPTION_GTID_BEGIN if we find a BEGIN.
Reset OPTION_GTID_BEGIN if we find a COMMIT.
sql/mysqld.cc:
Added slave_ddl_exec_mode_options
sql/mysqld.h:
Added slave_ddl_exec_mode_options
sql/rpl_gtid.cc:
Reset OPTION_GTID_BEGIN if we record a gtid (safety)
sql/sql_class.cc:
Regard all tables as transactional in commit if OPTION_GTID_BEGIN is set.
sql/sql_class.h:
Added to THD: log_current_statement and slave_expected_error
sql/sql_insert.cc:
Ensure that CREATE OR REPLACE is logged if table was deleted.
Don't do implicit commit for CREATE if we are under OPTION_GTID_BEGIN
sql/sql_parse.cc:
Change CREATE TABLE -> CREATE OR REPLACE TABLE for slaves
Change DROP TABLE -> DROP TABLE IF EXISTS for slaves
CREATE TABLE doesn't force implicit commit in case of OPTION_GTID_BEGIN
Don't do commits before or after any statement if OPTION_GTID_BEGIN was set.
sql/sql_priv.h:
Added OPTION_GTID_BEGIN
sql/sql_show.cc:
Enhanced store_create_info() to also be able to handle CREATE OR REPLACE
sql/sql_show.h:
Updated prototype
sql/sql_table.cc:
Ensure that CREATE OR REPLACE is logged if table was deleted.
sql/sys_vars.cc:
Added slave_ddl_exec_mode
sql/transaction.cc:
Added warning if we got a GTID under OPTION_GTID_BEGIN
on server shutdown after SELECT with CONVERT_TZ
It's wrong to return my_empty_string from val_str().
Removing my_empty_string. Using make_empty_result() instead.
Plugins get error notifications only when my_message_sql() is called.
But errors are launched with THD::raise_condition() calls in other
places. These are push_warning(), implementations of SIGNAL and
RESIGNAL commands.
So it makes sence to notify plugins there in THD::raise_condition().
Add another test case. This one for killing a query that is waiting
for a prior commit, when --log-slave-updates=0 (in this case the
wait happens in different code from --log-slave-updates=1).
Add a test case for killing a waiting query in parallel replication.
Fix several bugs found:
- We should not wakeup_subsequent_commits() in ha_rollback_trans(), since we
do not know the right wakeup_error() to give.
- When a wait_for_prior_commit() is killed, we must unregister from the
waitee so we do not race and get an extra (non-kill) wakeup.
- We need to deal with error propagation correctly in queue_for_group_commit
when one thread is killed.
- Fix one locking issue in queue_for_group_commit(), we could unlock the
waitee lock too early and this end up processing wakeup() with insufficient
locking.
- Fix Xid_log_event::do_apply_event; if commit fails it must not update the
in-memory @@gtid_slave_pos state.
- Fix and cleanup some things in the rpl_parallel.cc error handling.
- Add a missing check for killed in the slave sql driver thread, to avoid a
race.
