Date: Wed, 13 Oct 2021 15:52:46 -0300
From: Alvaro Herrera <alvherre@alvh.no-ip.org>
To: Andres Freund <andres@anarazel.de>
Cc: Tom Lane <tgl@sss.pgh.pa.us>, Andrew Dunstan <andrew@dunslane.net>,
	"Bossart, Nathan" <bossartn@amazon.com>,
	Kyotaro Horiguchi <horikyota.ntt@gmail.com>,
	"masao.fujii@oss.nttdata.com" <masao.fujii@oss.nttdata.com>,
	"pgsql-hackers@lists.postgresql.org" <pgsql-hackers@lists.postgresql.org>,
	"mengjuan.cmj@alibaba-inc.com" <mengjuan.cmj@alibaba-inc.com>,
	"Jakub.Wartak@tomtom.com" <Jakub.Wartak@tomtom.com>,
	Ryo Matsumura <matsumura.ryo@fujitsu.com>
Subject: Re: prevent immature WAL streaming
Message-ID: <202110131852.rrxabqmxxoiw@alvherre.pgsql>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <20211013183129.jutbvropp7nkg2kk@alap3.anarazel.de>
Archived-At: 
 <https://www.postgresql.org/message-id/202110131852.rrxabqmxxoiw%40alvherre.pgsql>
Precedence: bulk

On 2021-Oct-13, Andres Freund wrote:

> I added LSNs to the error message:
> not ok 1 - 000000010000000000000002 (0/2002350) differs from 000000010000000000000002 (0/2099600)
> 
> It appears that the problem is that inbetween the determination of
> rows_walsize the insert LSN moved to the next segment separately from the
> insertion, presumably due to autovacuum/analayze or such.
> <retries, with log_autovacuum_min_duration_statement=0, log_min_duration_statement=0>

Oh, of course.

> Hm. I guess we can disable autovac. But that's not a great solution, there
> might be WAL files due to catalog access etc too.

Well, we don't expect anything else to happen -- the cluster is
otherwise idle.  I think we should do it regardless of any other
changes, just to keep things steadier.

> Seems like it might be worth doing the "filling" of the segment with a loop in
> a DO block instead, where the end condition is to be within some distance to
> the end of the segment? With plenty headroom?

Eh, good idea ... didn't think of that, but it should keep things more
stable under strange conditions.

> Another thing: filling a segment by inserting lots of very tiny rows is pretty
> expensive. Can't we use something a bit wider? Perhaps even emit_message?

I think I realized partway through writing the test that I could use
emit_message instead of using a batched row insert ... so, yeah, we
can use it here also.

-- 
Álvaro Herrera              Valdivia, Chile  —  https://www.EnterpriseDB.com/
"La verdad no siempre es bonita, pero el hambre de ella sí"