Can still reproduce this, the description is unclear but in my case pm1 would show the warning and have no stats and pm2 is fine (which i think is what Daniel meant). Nothing obvious in logs.

This is tied up with cross engine join visibility. If you change the cross engine join config to use the pm1 hostname then pm1 works but now pm2 no longer works.

Further there seems to be an error condition when the insert into querystats is performed from a remote host, e.g. pm2 exemgr:

> insert delayed into querystats values (0, 123, "centos2", "root", "1", "ABC", "select 1 from foo","2017-01-01", "2017-01-01",1,123, 0, 0, 1,2,1,2,3,4,5,6);

ERROR 1616 (HY000): DELAYED option not supported for table 'querystats'

One question is why does a query executed on um/pm 1 always want to use the exemgr on pm2? It would seem more efficient to use the local one and then a localhost cross engine join config would work.

The system does round robin between exemgrs and this is also broken in infinidb. This will require some architectural changes so is more than just a small maintenance release bug fix. Essentially QueryStats will only work fully in an environment with one UM. It will partially record on a multi um deployment.

The actual root cause is due to a porting issue to mariadb. The querystats table was updated to use InnoDB but the querystats insert statement uses the insert delayed syntax which is not supported by InnoDB. Thus the fix for this is to either revert to using a MyISAM table or update the insert statement to remove the delayed keyword.

A workaround for now is to migrate the querystats table to use MyISAM after installation:

 alter table infinidb_querystats.querystats engine=MyISAM;

The change for develop was mistakenly directly cherry-picked in, so there's no pull request for that, only for develop-1.0

Builds tested: 1.0.1-1 and 1.1.0-1 (GitHub source build)

1.0.1-1
On a combo stack (PM1 and PM2), DML sync is one way, from PM1 to PM2. With querystats enabled, when executing a query on PM1, querystats on both PM1 and PM2 gets a new row. This is the expected behavior. When executing a query on PM2, only querystats on PM2 gets the new row. With the current implementation, this is to be expected. But I don't think this is the EXPECTED behavior.

1.1.0

When querystats is enabled, mysqld crashes on any user query.

MariaDB [mytest]> select count from orders;
ERROR 2013 (HY000): Lost connection to MySQL server during query

No entries recorded in err.log and crit.log. mysqld gets a new PID. This issue occurs on both PMs.

reopen per my last comment

Please open a new bug for 1.1.0 and then we can remove the fixed in here.

For 1.0.x this is the correct behavior based on having the default setup of replication on and using localhost as the cross engine join host.

• To have the querystats on all hosts, configure the cross engine join host as UM1 which is the replication master
• To have the querystats on each host, configure cross engine join as localhost (also exemgr hosts too) and either disable replication entirely or just for the querystats db / table: https://mariadb.com/kb/en/mariadb/replication-and-binary-log-server-system-variables/#replicate_ignore_db

opened ticket MCOL-879 to track the mysqld crashing issue. Closing the ticket for 1.0.11-1