On Sun, Oct 24, 2021 at 4:57 AM Andrew Gierth
<andrew@tao11.riddles.org.uk> wrote:
> CC'ing to original reporter.
I suggest that pg_amcheck be run against the database, to try to
establish the overall extent of the corruption -- including all
indexes and heap relations from the database. This should start with
the original known-corrupt backup, from before the REINDEX. Then
report back the output of pg_amcheck to the list.
See: https://www.postgresql.org/docs/14/app-pgamcheck.html
Please specify --heapallindexed and --parent-check options when
running pg_amcheck, so that it will perform the most thorough possible
checks against indexes. --heapallindexed is particularly important
here, because the issue that Andrew highlighted is probably not
detectable just by looking at the index itself. It's purely an
inconsistency between the index and the table. This class of problem
has many possible causes.
Thanks
--
Peter Geoghegan