Horton

Re: logical replication: restart_lsn can go backwards (and more), seems broken since 9.4

Michael Paquier <michael@paquier.xyz>

View thread View on postgr.es
From: Michael Paquier <michael@paquier.xyz>
To: Ashutosh Bapat <ashutosh.bapat.oss@gmail.com>
Cc: Tomas Vondra <tomas@vondra.me>, Masahiko Sawada <sawada.mshk@gmail.com>, PostgreSQL Hackers <pgsql-hackers@postgresql.org>
Date: 2024-11-14T06:34:11Z
Lists: pgsql-hackers 
On Thu, Nov 14, 2024 at 11:45:56AM +0530, Ashutosh Bapat wrote:
> Here's a quick and dirty patch which describes the idea. I didn't get
> time to implement code to move SnapBuild::restart_lsn if
> SnapBuild::start_decoding_at moves forward while building initial
> snapshot. I am not sure whether that's necessary either.
> 
> I have added three elogs to see if the logic is working as expected. I
> see two of the elogs in patch in the server log when I run tests from
> tests/subscription and tests/recovery. But I do not see the third one.
> That either means that the situation causing the bug is not covered by
> those tests or the fix is not triggered. If you run your reproduction
> and still see the crashes please provide the output of those elog
> messages along with the rest of the elogs you have added.

Forgot the attachment, perhaps?
--
Michael

Commits

  1. Fix a possibility of logical replication slot's restart_lsn going backwards.

  2. Introduce logical decoding.