Re: [PATCH] Disable bgworkers during servers start in pg_upgrade

From: Michael Paquier <michael(at)paquier(dot)xyz>
To: Julien Rouhaud <rjuju123(at)gmail(dot)com>
Cc: Bruce Momjian <bruce(at)momjian(dot)us>, Daniel Gustafsson <daniel(at)yesql(dot)se>, Denis Laxalde <denis(dot)laxalde(at)dalibo(dot)com>, Jehan-Guillaume de Rorthais <jgdr(at)dalibo(dot)com>, Andres Freund <andres(at)anarazel(dot)de>, PostgreSQL Hackers <pgsql-hackers(at)lists(dot)postgresql(dot)org>
Subject: Re: [PATCH] Disable bgworkers during servers start in pg_upgrade
Date: 2021-08-27 04:41:39
Message-ID: YShtA86sOfazb9at@paquier.xyz
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On Fri, Aug 27, 2021 at 11:25:19AM +0800, Julien Rouhaud wrote:
> Right now AFAICT there's no official API to check if a call to
> RegisterBackgroundWorker() succeeded or not, but an extension could
> check by itself using BackgroundWorkerList in bgworker_internals.h,
> and error out or something if it didn't succeed, as a way to inform
> users that they didn't configure the instance properly (like maybe
> increasing max_worker_processes). Surely using a *_internals.h header
> is a clear sign that you expose yourself to problems, but adding an
> official way to check for bgworker registration doesn't seem
> unreasonable to me. Is that worth the risk to have pg_upgrade
> erroring out in this kind of scenario, or make the addition of a new
> API to check for registration status more difficult?

Perhaps. That feels like a topic different than what's discussed
here, though, because we don't really need to check if a bgworker has
been launched or not. We just need to make sure that it never runs in
the context of a binary upgrade, like autovacuum.
--
Michael

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Greg Nancarrow 2021-08-27 04:55:34 Fix erroneous parallel execution when modifying CTE is present in rewritten query
Previous Message Amit Kapila 2021-08-27 04:36:48 Re: Skipping logical replication transactions on subscriber side