From: Tom Lane <tgl@sss.pgh.pa.us>
To: Andres Freund <andres@anarazel.de>
cc: pgsql-hackers@lists.postgresql.org, Noah Misch <noah@leadboat.com>,
        Peter Geoghegan <pg@bowt.ie>, David Rowley <dgrowleyml@gmail.com>
Subject: Re: Intermittent buildfarm failures on wrasse
In-reply-to: <A11045E8-18D2-4BBB-8CF1-80DC6920ED73@anarazel.de>
References: <1521388.1649953580@sss.pgh.pa.us>
 <1547912.1649971995@sss.pgh.pa.us>
 <CAH2-Wz=DHY2LV7=2aVd=qz6AbQhFK7KJRNL5ZwgB7dyno_gzdg@mail.gmail.com>
 <1553951.1649975018@sss.pgh.pa.us>
 <CAH2-WzngTjUk_65ESmrMf40WTqasrmWG0gpwnvcN+mXsoXT7qQ@mail.gmail.com>
 <CAH2-WznXXkkfE_y9ieaHkEQj31HeMAqJBtJPmDY+EpyVg1QywQ@mail.gmail.com>
 <1570894.1649986347@sss.pgh.pa.us>
 <20220415015249.kfsqxujrf2bn2wz6@alap3.anarazel.de>
 <20220415024515.GA862547@rfd.leadboat.com>
 <20220415051205.6abhlilh6fnjgohr@alap3.anarazel.de>
 <20220415052125.GD862547@rfd.leadboat.com> <1637671.1650032132@sss.pgh.pa.us>
 <1643686.1650036220@sss.pgh.pa.us>
 <A11045E8-18D2-4BBB-8CF1-80DC6920ED73@anarazel.de>
Comments: In-reply-to Andres Freund <andres@anarazel.de>
	message dated "Fri, 15 Apr 2022 12:22:41 -0400"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <1650515.1650040612.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Fri, 15 Apr 2022 12:36:52 -0400
Message-ID: <1650516.1650040612@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/1650516.1650040612%40sss.pgh.pa.us>
Precedence: bulk

Andres Freund <andres@anarazel.de> writes:
> On April 15, 2022 11:23:40 AM EDT, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> The something is the logical replication launcher.  In the failing runs=
,
>> it is advertising xmin =3D 724 (the post-initdb NextXID) and continues =
to
>> do so well past the point where tenk1 gets vacuumed.

> That explains it. Before shmstat autovac needed to wait for the stats co=
llector to write out stats. Now it's near instantaneous. So the issue prob=
ably existed before, just unlikely to ever be reached.

Um, this is the logical replication launcher, not the autovac launcher.
Your observation that a sleep in get_database_list() reproduces it
confirms that, and I don't entirely see why the timing of the LR launcher
would have changed.

(On thinking about it, I suppose the AV launcher might trigger this
too, but that is not the PID I saw in testing.)

> We can't just ignore database less xmins for non-shared rels, because wa=
lsender propagates hot_standby_feedback that way. But we can probably add =
a flag somewhere indicating whether a database less PGPROC has to be accou=
nted in the horizon for non-shared rels.

Yeah, I was also thinking about a flag in PGPROC being a more reliable
way to do this.  Is there anything besides walsenders that should set
that flag?

			regards, tom lane