Re: To what extent should tests rely on VACUUM ANALYZE?

On Thu, Mar 28, 2024 at 11:00 PM Alexander Lakhin <exclusion(at)gmail(dot)com>
wrote:

> When running multiple 027_stream_regress.pl test instances in parallel
> (and with aggressive autovacuum) on a rather slow machine, I encountered
> test failures due to the subselect test instability just as the following
> failures on buildfarm:
> 1)
> https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=grassquit&dt=2024-03-27%2010%3A16%3A12
>
> ---
> /home/bf/bf-build/grassquit/HEAD/pgsql/src/test/regress/expected/subselect.out
> 2024-03-19 22:20:34.435867114 +0000
> +++
> /home/bf/bf-build/grassquit/HEAD/pgsql.build/testrun/recovery/027_stream_regress/data/results/subselect.out
>
> 2024-03-27 10:28:38.185776605 +0000
> @@ -2067,16 +2067,16 @@
> QUERY PLAN
> -------------------------------------------------
> Hash Join
> - Hash Cond: (c.odd = b.odd)
> + Hash Cond: (c.hundred = a.hundred)
> -> Hash Join
> - Hash Cond: (a.hundred = c.hundred)
> - -> Seq Scan on tenk1 a
> + Hash Cond: (b.odd = c.odd)
> + -> Seq Scan on tenk2 b
> -> Hash
> -> HashAggregate
> Group Key: c.odd, c.hundred
> -> Seq Scan on tenk2 c
> -> Hash
> - -> Seq Scan on tenk2 b
> + -> Seq Scan on tenk1 a
> (11 rows)

FWIW, this issue is also being reproduced in Cirrus CI, as Matthias
reported in another thread [1] days ago.

[1]
https://www.postgresql.org/message-id/CAEze2WiiE-iTKxgWQzcjyiiiA4q-zsdkkAdCaD_E83xA2g2BLA%40mail.gmail.com

Thanks
Richard