Re: Offline data checksum changes can cause incorrect checksum state on standbys

From: Zsolt Parragi <zsolt(dot)parragi(at)percona(dot)com>
To: Heikki Linnakangas <hlinnaka(at)iki(dot)fi>
Cc: pgsql-hackers(at)lists(dot)postgresql(dot)org, Daniel Gustafsson <daniel(at)yesql(dot)se>, Bertrand Drouvot <bertranddrouvot(dot)pg(at)gmail(dot)com>
Subject: Re: Offline data checksum changes can cause incorrect checksum state on standbys
Date: 2026-09-08 21:21:42
Message-ID: CAN4CZFPnT-VcrQ-jqHNCMvfObmsbZTOxrmgW1F1L1rRscu2xnA@mail.gmail.com
Views: Whole Thread | Raw Message | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On Tue, 08 Sep 2026, Heikki Linnakangas
<heikki(dot)linnakangas(at)enterprisedb(dot)com> wrote:
> In principle, I think pg_rewind would still work as long as
> wal_log_hints=on. But I don't think we need to cater for that, erroring
> out is fine.

The problem with rewind/backups/etc in a mixed cluster is all the
corner cases. Everything should work, except that we can end up with
checksum errors because an instance that supposedly has checksums
enabled ends up with a few pages without checksums. Most of the time
the issue is immediately visible, but we can construct scenarios where
users only get those checksum errors much later, when it is no longer
clear if its a real corruption or a leftover issue from some tool use.

In response to

Browse pgsql-hackers by date

  From Date Subject
Next Message Daniel Gustafsson 2026-09-08 21:38:19 Trying to break online checksums with LLMs
Previous Message Masahiko Sawada 2026-09-08 21:12:44 Re: REPACK (CONCURRENTLY) can crash a logical decoding session