On Mon, Apr 6, 2015 at 07:21:30PM +0000, tgarnett@panjiva.com wrote:
> We've retained the pre-failure snapshot and wal files for now (though we
> will need to free them up at some point) so we can point in time recover to
> any point from 28 hours before to several hours after the problem surfaced
> if that's helpful at all.
>
> Bugs / Questions:
> - the members files wrapped over themselves leading to corruption
> - why didn't vacuum_db -a -F free up more members files?
Wow, you did a lot of research on this and I am glad you seem to have
found a solution. Frankly, there were so many multi-xact bugs in 9.3.X
that I am unable to track exactly what bugs caused what failures, when
they were fixed, and whether fixes fixed the problem for good, or
whether they just fixed them from happening in the future. Yes, very
discouraging.
--
Bruce Momjian <bruce@momjian.us> http://momjian.us
EnterpriseDB http://enterprisedb.com
+ Everyone has their own god. +