Re: visibility map corruption

On Sun, Jul 4, 2021 at 10:28:25PM +0000, Floris Van Nee wrote:
> >
> > I wonder if it's related to this issue:
> >
> > https://www.postgresql.org/message-
> > id/20210423234256(dot)hwopuftipdmp3okf(at)alap3(dot)anarazel(dot)de
> >
> > Have you increased autovacuum_freeze_max_age from its default? This
> > already sounds like the kind of database where that would make
> > sense.
> >
>
> autovacuum_freeze_max_age is increased in our setup indeed (it is
> set to 500M). However, we do regularly run manual VACUUM (FREEZE)
> on individual tables in the database, including this one. A lot of
> tables in the database follow an INSERT-only pattern and since it's
> not running v13 yet, this meant that these tables would only rarely
> be touched by autovacuum. Autovacuum would sometimes kick in on some
> of these tables at the same time at unfortunate moments. Therefore we
> have some regular job running that VACUUM (FREEZE)s tables with a xact
> age higher than a (low, 10M) threshold ourselves.

OK, this is confirmation that the pg_resetwal bug, and its use by
pg_upgrade, is a serious issue that needs to be addressed. I am
prepared to work on it now.

--
Bruce Momjian <bruce(at)momjian(dot)us> https://momjian.us
EDB https://enterprisedb.com

If only the physical world exists, free will is an illusion.