Comment 13 for bug 1366997

Hi Mario,

I'm fairly certain that issuing FLUSH TABLES on all nodes in the cluster is also a workaround to fix these errors after they have started. It won't prevent the problem from occurring, but it might be useful if you need to keep wsrep_auto_increment_control ON and you don't use autocommit.

My theory (somewhat validated by the 5.6 patch) is that the problem corrects itself with the next statement using a cached table. That could take a while if you have a large table cache setting. FLUSH TABLES will therefore clear that cache and speed the process along.