Online GTID gtid_deployment_step replication discrepancy

[monty solomon]

I used gtid_deployment_step to convert multiple clusters to use GTID. After completing the conversions I ran pt-table-checksum and discovered that there were replication inconsistencies on a bunch of the clusters. Some of the changes on the new master were not replicated to the slaves.

The replication discrepancies appeared on clusters that were very busy. It appears that there may be a window during the process where the new master is writing binlog entries without GTIDs.

When exactly during the master/slave swap should the value of gtid_deployment_step be changed on the slave that will become the new master?

Online GTID deployment
https://www.percona.com/doc/percona-server/5.6/flexibility/online_gtid_deployment.html

Online GTID rollout now available in Percona Server 5.6
https://www.percona.com/blog/2015/02/10/online-gtid-rollout-now-available-percona-server-5-6/

[Laurynas Biveinis (laurynas-biveinis)]

Monty, we have just learned that gtid_deployment_step would release any FTWRL on that connection (bug 1549229). Do you use FTWRL here? Would that help to explain your symptoms?

[monty solomon (monty+launchpad)]

That does bug not seem to be related.

[monty solomon (monty+launchpad)]

(Launchpad needs a feature to edit comments :-) )

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-3374