Re: Wrong results with postgres_fdw and merge anti join from RHEL 7.9 to RHEL 8.7

Daniel Westermann (DWE) <daniel.westermann@dbi-services.com>

From: "Daniel Westermann (DWE)" <daniel.westermann@dbi-services.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: "pgsql-bugs@lists.postgresql.org" <pgsql-bugs@lists.postgresql.org>
Date: 2023-04-05T19:20:21Z
Lists: pgsql-bugs
>Yeah, doesn't look like you've made any configuration mistakes.
>So either the two OSes sort differently, or there's index corruption
>causing the indexscan to give bogus output.

We did a rebuild of the indexes and also vacuum full, just to be sure. Did not change anything.

>The sample data you showed seemed to only involve numeric-ish strings,
>which would be highly unlikely to change sort order across locale
>updates.  But maybe there are weirder entries elsewhere in the column?

I can probably provide a dump, but I've to ask. Would that help?

>Anyway, the first thing I'd try is reindexing both tables --- doesn't
>look like they're large enough to make that painful.  If that doesn't
>fix it you must have a collation difference.  (Asking both systems
>for a sorted dump of their cprd columns could help confirm that.)
>You could probably hack around that, if an OS update isn't feasible,
>by labelling the foreign table's column with some collation you aren't
>using anywhere else in the local database.

I'll try do that tomorrow.

Thanks
Daniel