specific temporary errors
The optimistic parallel slave's worker thread could face a run-time error due to
the algorithm's specifics which allows for conflicts like the reported
"Can't find record in 'table'".
A typical stack is like
{noformat}
#0 handler::print_error (this=0x61c00008f8a0, error=149, errflag=0) at handler.cc:3650
#1 0x0000555555e95361 in write_record (thd=thd@entry=0x62a0000a2208, table=table@entry=0x61f00008ce88, info=info@entry=0x7fffdee356d0) at sql_insert.cc:1944
#2 0x0000555555ea7767 in mysql_insert (thd=thd@entry=0x62a0000a2208, table_list=0x61b00012ada0, fields=..., values_list=..., update_fields=..., update_values=..., duplic=<optimized out>, ignore=<optimized out>) at sql_insert.cc:1039
#3 0x0000555555efda90 in mysql_execute_command (thd=thd@entry=0x62a0000a2208) at sql_parse.cc:3927
#4 0x0000555555f0cc50 in mysql_parse (thd=0x62a0000a2208, rawbuf=<optimized out>, length=<optimized out>, parser_state=<optimized out>) at sql_parse.cc:7449
#5 0x00005555566d4444 in Query_log_event::do_apply_event (this=0x61200005b9c8, rgi=<optimized out>, query_arg=<optimized out>, q_len_arg=<optimized out>) at log_event.cc:4508
#6 0x00005555566d639e in Query_log_event::do_apply_event (this=<optimized out>, rgi=<optimized out>) at log_event.cc:4185
#7 0x0000555555d738cf in Log_event::apply_event (rgi=0x61d0001ea080, this=0x61200005b9c8) at log_event.h:1343
#8 apply_event_and_update_pos_apply (ev=ev@entry=0x61200005b9c8, thd=thd@entry=0x62a0000a2208, rgi=rgi@entry=0x61d0001ea080, reason=<optimized out>) at slave.cc:3479
#9 0x0000555555d8596b in apply_event_and_update_pos_for_parallel (ev=ev@entry=0x61200005b9c8, thd=thd@entry=0x62a0000a2208, rgi=rgi@entry=0x61d0001ea080) at slave.cc:3623
#10 0x00005555562aca83 in rpt_handle_event (qev=qev@entry=0x6190000fa088, rpt=rpt@entry=0x62200002bd68) at rpl_parallel.cc:50
#11 0x00005555562bd04e in handle_rpl_parallel_thread (arg=arg@entry=0x62200002bd68) at rpl_parallel.cc:1258
{noformat}
Here {{handler::print_error}} computes whether to error log the
current error when --log-warnings > 1. The decision flag is consulted
bu {{my_message_sql()}} which can be eventually called.
In the bug case the decision is to log.
However in the optimistic mode slave applier case any conflict is
attempted to resolve with rollback and retry to success. Hence the
logging is at least extraneous.
The case is fixed with adding a new flag {{ME_LOG_AS_WARN}} which
{{handler::print_error}} may propagate further on through {{my_error}}
when the error comes from an optimistically running slave worker thread.
The new flag effectively requests the warning level for the errlog record,
while the thread's DA records the actual error (which is regarded as temporary one
by the parallel slave error handler).
These test can sporadically show mutex deadlock warnings between LOCK_wsrep_thd
and LOCK_thd_data mutexes. This means that these mutexes can be locked in opposite
order by different threads, and thus result in deadlock situation.
To fix such issue, the locking policy of these mutexes should be revised and
enforced to be uniform. However, a quick code review shows that the number of
lock/unlock operations for these mutexes combined is between 100-200, and all these
mutex invocations should be checked/fixed.
On the other hand, it turns out that LOCK_wsrep_thd is used for protecting access to
wsrep variables of THD (wsrep_conflict_state, wsrep_query_state), whereas LOCK_thd_data
protects query, db and mysys_var variables in THD. Extending LOCK_thd_data to protect
also wsrep variables looks like a viable solution, as there should not be a use case
where separate threads need simultaneous access to wsrep variables and THD data variables.
In this commit LOCK_wsrep_thd mutex is refactored to be replaced by LOCK_thd_data.
By bluntly replacing LOCK_wsrep_thd by LOCK_thd_data, will result in double locking
of LOCK_thd_data, and some adjustements have been performed to fix such situations.
ALTER TABLE ... ADD PARTITION modifies the open TABLE structure,
and sets table->need_reopen=1 to reset these modifications
in case of an error.
But under LOCK TABLES the table isn't get reopened, despite need_reopen.
Fixed by reopening need_reopen tables under LOCK TABLE.
Counter for select numbering made stored with the statement (before was global)
So now it does have always accurate value which does not depend on
interruption of statement prepare by errors like lack of table in
a view definition.
The function trans_rollback_to_savepoint(), unlike trans_savepoint(),
did not allow xa_state=XA_ACTIVE, so an attempt to do ROLLBCK TO SAVEPOINT
inside an XA transaction incorrectly returned an error
"...command cannot be executed ... in the ACTIVE state...".
Partially merging a MySQL patch:
7fb5c47390311d9b1b5367f97cb8fedd4102dd05
This is WL#7193 (Decouple THD and st_transactions)...
The currently merged part includes these changes:
- Introducing st_xid_state::check_has_uncommitted_xa()
- Reusing it in both trans_rollback_to_savepoint() and trans_savepoint(),
so now both allow XA_ACTIVE.
followup for 97c2a7354b - don't use thd->is_error(),
the error could've been set before TABLE_LIST::cleanup_items.
Use the error handler to count errors.
This fixes rpl.rpl_row_binlog_max_cache_size - it was failing when
ER_STMT_CACHE_FULL happened duing multi-table update. Because
multi_update::abort_result_set() calls do_updates() to update
as much as possible, so one cannot rely on thd->is_error() after that.
Fixed by making sure that the sort buffer would have atleast MERGEBUFF2 keys.
Also fixed MDEV-13457 by making sure that an empty tree is never dumped to the disk
The problem was that the introduction of max-thread-mem-used can cause
an allocation error very early, even before mysql_parse() is called.
As mysql_parse() calls thd->reset_for_next_command(), which called
clear_error(), the error number was lost.
Fixed by adding an option to have unique messages for each KILL
signal and change max-thread-mem-used to use this new feature.
This removes a lot of problems with the original approach, where
one could get errors signaled silenty almost any time.
ixed by moving clear_error() from reset_for_next_command() to
do_command(), before any memory allocation for the thread.
Related changes:
- reset_for_next_command() now have an optional parameter if we should
call clear_error() or not. By default it's called, but not anymore from
dispatch_command() which was the original problem.
- Added optional paramater to clear_error() to force calling of
reset_diagnostics_area(). Before clear_error() only called
reset_diagnostics_area() if there was no error, so we normally
called reset_diagnostics_area() twice.
- This change removed several duplicated calls to clear_error()
when starting a query.
- Reset max_mem_used on COM_QUIT, to protect against kill during
quit.
- Use fatal_error() instead of setting is_fatal_error (cleanup)
- Set fatal_error if max_thead_mem_used is signaled.
(Same logic we use for other places where we are out of resources)
CREATE/DROP TEMPORARY TABLE are not safe to optimistically replicate in
parallel with other transactions, so they need to be marked as "ddl" in the
binlog.
This was already done for stand-alone CREATE/DROP TEMPORARY. But temporary
tables can also be created and dropped inside a BEGIN...END transaction, and
such transactions were not marked as ddl. Nor was the DROP TEMPORARY TABLE
statement emitted implicitly when a client connection is closed.
So this patch adds such ddl mark for the missing cases.
The difference to Kristian's original patch is mainly a fix in
mysql_trans_commit_alter_copy_data() to remember the unsafe_rollback_flags
over the temporary commit.
it used current_thd->alloc() and allocated on the thd's execution arena,
not on table->expr_arena.
Remove THD::arena_for_cached_items that is temporarily set in
update_virtual_fields(), and replaces THD arena in get_datetime_value().
Instead set THD arena to table->expr_arena for the whole duration
of update_virtual_fields()
When a deadlock kill is detected inside the storage engine, the kill
is not done immediately, to avoid calling back into the storage engine
kill_query method with various lock subsystem mutexes held. Instead the
kill is queued and done later by a slave background thread.
This patch in preparation for fixing TokuDB optimistic parallel
replication, as well as for removing locking hacks in InnoDB/XtraDB in
10.2.
Signed-off-by: Kristian Nielsen <knielsen at knielsen-hq.org>
- reverted from tracking donor servicing thread. With xtrabackup SST,
xtrabackup thread will call FTWRL and node is desynced upfront
- Skipping desync in FTWRL if node is operating as donor
- enveloped FTWRL processing with wsrep desync/resync calls. This way FTWRL processing node
will not cause flow control to kick in
- donor servicing thread is unfortunate exception, we must let him to pause provider as part
of FTWRL phase, but not desync/resync as this is done as part of donor control on higher
level
- Fixed typos
- Added --core-on-failure to mysql-test-run
- More DBUG_PRINT in viosocket.c
- Don't forget CLIENT_REMEMBER_OPTIONS for compressed slave protocol
- Removed not used stage variables
(like DROP TABLE) has been scheduled before conflicting DDL's (like INSERT)
are commited.
What makes these bugs hard to detect is that in most cases any wrong
schduling are caught by MDL locks. It's only when there are timing issues
that the bugs (usually deadlocks) are noticed.
This patch adds a DBUG_ASSERT() that detects, in parallel replication,
if a DDL is scheduled before any depending DML'S are commited.
It does this be checking if there are any conflicting replication locks
when the DDL is about to wait for getting it's MDL lock.
I also did some minor code cleanups in sql_base.cc to make this code
similar to other related code.
Transaction replay causes the THD to re-apply the replication
events from execution, using the same path appliers do. While
applying the log events, the THD's timestamp is set to the
timestamp of the event.
Setting the timestamp explicitly causes function NOW() to
always the timestamp that was set. To avoid this behavior we
reset the timestamp after replaying is done.
