On Tue, Oct 6, 2015 at 2:10 AM, Jeff Janes wrote:
> I was assuming that the table being AccessExclusive locked and the table
> being VACUUMed were two different tables, which were done in two different
> processes on the master, which means that no re-ordering of WAL records
> would be necessary to hit this.
Right. This may be possible as well.
> Although I also haven't yet been able to reproduce the deadlock under this assumption.
Yeah, we are a bit blind here. Maxim, did you perform an operation on
master that took an exclusive lock on the relation involved here? Was
this relation a table, an index? If an index, what kind of index? I
think that such information may help us narrowing down a test case.
--
Michael