Heikki Linnakangas <hlinnaka@iki.fi> writes:
> I was finally able to reproduce this, by running the failing pg_regress
> tests concurrently with repeated "vacuum full pg_database" calls. It's
> curious that 'prion' failed on the first run after the commit, I was not
> able to reproduce it by just running the unmodified regression tests
> with -DRELCACHE_FORCE_RELEASE -DCATCACHE_FORCE_RELEASE.
> Committed a fix. There was one codepath that was missing a call to
> RelationInitPhysicalAddr(relation) after the patch.
Just as a note, BF members broadbill and mamushi showed similar
symptoms [1][2]. Neither of them use FORCE_RELEASE options AFAICS.
I'm guessing appearance of the failure is simply a timing issue.
regards, tom lane
[1] https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=broadbill&dt=2024-10-31%2013%3A34%3A08
[2] https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=mamushi&dt=2024-10-31%2016%3A08%3A43
