Comment 2 for bug 1257069

This is a conceptual problem with wsrep replication model caused by the need to control the order in which results of DDL operation become effective (when DDL operation ends). From wsrep perspective it is then that the database state undergoes the atomic change. And this change must happen in the same order on all nodes - no sooner, no later. So far MySQL implementation of DDL operations does not provide any means to control when they become effective. Since we have no way to control this, we need to pause execution of any other operations until current DDL ends.

In other words, what is important is when DDL operation ends, and for MySQL it is a matter of luck. Contrast this to transaction execution, for which we have COMMIT.