From: Antonin Houska <ah@cybertec.at>
To: Arseny Sher <a.sher@postgrespro.ru>
cc: pgsql-hackers <pgsql-hackers@postgresql.org>
Subject: Re: Fix slot's xmin advancement and subxact's lost snapshots in
 decoding.
In-reply-to: <87lgdyz1wj.fsf@ars-thinkpad>
References: <87lgdyz1wj.fsf@ars-thinkpad>
Comments: In-reply-to Arseny Sher <a.sher@postgrespro.ru>
   message dated "Sun, 08 Apr 2018 08:46:04 +0300."
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
Date: Mon, 11 Jun 2018 09:07:47 +0200
Message-ID: <15387.1528700867@localhost>
Precedence: bulk

Arseny Sher <a.sher@postgrespro.ru> wrote:
> Please see detailed description of the issues, tests which reproduce
> them and fixes in the attached patch.

I've played with your tests and gdb for a while, both w/o and with your
patch. I think I can understand both problems. I could not invent simpler w=
ay
to fix them.

One comment about the coding: since you introduce a new transaction list and
it's sorted by LSN, I think you should make the function AssertTXNLsnOrder()
more generic and use it to check the by_base_snapshot_lsn list too. For
example call it after the list insertion and deletion in
ReorderBufferPassBaseSnapshot().

Also I think it makes no harm if you split the diff into two, although both
fixes use the by_base_snapshot_lsn field.

--=20
Antonin Houska
Cybertec Sch=C3=B6nig & Sch=C3=B6nig GmbH
Gr=C3=B6hrm=C3=BChlgasse 26, A-2700 Wiener Neustadt
Web: https://www.cybertec-postgresql.com