Description:
I encounter an issue in our production environment.
Although set slave_preserve_commit_order=1 in master/slave instance, We still encounter gtid disorder in Slave
In master's binlog we find the gtid sequence
15959d02-136e-11e9-a1e5-ac751d25dbc3:68690512：last_committed=1143 sequence_number=1144
15959d02-136e-11e9-a1e5-ac751d25dbc3:68690513 ：last_committed=1144 sequence_number=1145
15959d02-136e-11e9-a1e5-ac751d25dbc3:68690514：last_committed=1144 sequence_number=1146
In slave's binlog
15959d02-136e-11e9-a1e5-ac751d25dbc3:68690512：last_committed=1143 sequence_number=1144
15959d02-136e-11e9-a1e5-ac751d25dbc3:68690514：last_committed=1144 sequence_number=1145
15959d02-136e-11e9-a1e5-ac751d25dbc3:68690513：last_committed=1144 sequence_number=1146
So
slave's gtid
<uuid>:68690514 ---> <uuid>:68690513
maste's gtid
<uuid>:68690513 ---> <uuid>:68690514
After several day's investigation, we find some MyIsam table in out schema.
my.cnf
slave_pending_jobs_size_max=16M
binlog_order_commits=1
slave_preserve_commit_order=1
slave_parallel_workers=12
slave_parallel_type=logical_clock
How to repeat:
my.cnf
slave_pending_jobs_size_max=16M
binlog_order_commits=1
slave_preserve_commit_order=1
slave_parallel_workers=12
slave_parallel_type=logical_clock
0. Construct a Master / Slave cluster, Make sure using the configuration mentioned above
1.create a table using MyIsam Engine.
2.Insert some data in parallel(Make group commit take effect) .
3. compare the gtid sequence between master and slave.
4. repeat the 2nd step until you find gtid disorder
Suggested fix:
I suggest we should give a warning in error log when the following condition holds
1. slave_parallel_workers >0
2.slave_parallel_type=logical_clock
3.slave_preserve_commit_order=1
4. the binlog event applying is from a Myisam table
