Horton

Re: wake up logical workers after ALTER SUBSCRIPTION

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Nathan Bossart <nathandbossart@gmail.com>
Cc: Melih Mutlu <m.melihmutlu@gmail.com>, Thomas Munro <thomas.munro@gmail.com>, "Hayato Kuroda (Fujitsu)" <kuroda.hayato@fujitsu.com>, "pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>
Date: 2022-12-14T19:02:58Z
Lists: pgsql-hackers 
Nathan Bossart <nathandbossart@gmail.com> writes:
> On Wed, Dec 14, 2022 at 01:23:18PM -0500, Tom Lane wrote:
>> Oh.  What in the world is the rationale for that?

> My assumption is that this is meant to avoid starting workers as fast as
> possible if they repeatedly crash.

I can see the point of rate-limiting if the workers are failing to connect
or crashing while trying to process data.  But it's not very sane to
apply the same policy to an intentional worker exit-for-reconfiguration.

Maybe we could have workers that are exiting for that reason set a
flag saying "please restart me without delay"?

A *real* fix would be to not exit at all, at least for reconfigurations
that don't change the connection parameters, but instead cope with
recomputing whatever needs recomputed in the workers' state.  I can
believe that that'd be a lot of work though.

			regards, tom lane

Commits

  1. Avoid type cheats for invalid dsa_handles and dshash_table_handles.

  2. Track logrep apply workers' last start times to avoid useless waits.

  3. Check for two_phase change at end of process_syncing_tables_for_apply.

  4. Wake up a subscription's replication worker processes after DDL.