Andres Freund <andres(at)2ndquadrant(dot)com> writes:
> The VACUUM implementation in 9.3 had several bugs: It removed multixact
> xmax values without regard of the importance of contained xids, it did
> not remove multixacts if the contained xids were too old and it relied
> on hint bits when checking whether a row needed to be frozen which might
> not have been set on replicas.
Uh ... what does the last have to do with it? Surely we don't run
VACUUM on replicas. Or are you talking about what might happen when
VACUUM is run on a former replica that's been promoted to master?
regards, tom lane