Date: Wed, 29 Nov 2023 11:52:01 -0600
From: Nathan Bossart <nathandbossart@gmail.com>
To: Andres Freund <andres@anarazel.de>
Cc: David Geier <geidav.pg@gmail.com>,
	PostgreSQL-development <pgsql-hackers@postgresql.org>
Subject: Re: Fix assertion in autovacuum worker
Message-ID: <20231129175201.GA902323@nathanxps13>
References: <b12f9e22-2618-42b8-8644-88bae192c7fd@gmail.com>
 <20231128220516.GA475168@nathanxps13>
 <20231129000349.pcfvrewvqk4he6i2@awork3.anarazel.de>
 <20231129024247.GA479372@nathanxps13>
 <20231129024859.kajlgqg33dksonbi@awork3.anarazel.de>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20231129024859.kajlgqg33dksonbi@awork3.anarazel.de>
Archived-At: <https://www.postgresql.org/message-id/20231129175201.GA902323%40nathanxps13>
Precedence: bulk

On Tue, Nov 28, 2023 at 06:48:59PM -0800, Andres Freund wrote:
> On 2023-11-28 20:42:47 -0600, Nathan Bossart wrote:
>> Right.  Perhaps we could add a LWLockReleaseAll() to
>> pgstat_shutdown_hook() instead of the autovacuum code, but I'm afraid that
>> is still just a hack.
> 
> Yea, we'd need that in just about all before_shmem_exit() callbacks.  I could
> see an argument for doing it in proc_exit_prepare(). While that'd be a fairly
> gross layering violation, we already do reset a number a bunch of stuff in
> there:

Gross layering violations aside, that at least seems more future-proof
against other sigsetjmp() blocks that proc_exit() without doing any
preliminary cleanup.

-- 
Nathan Bossart
Amazon Web Services: https://aws.amazon.com