From: Tom Lane <tgl@sss.pgh.pa.us>
To: jmv@emorozov.net
cc: pgsql-bugs@lists.postgresql.org
Subject: Re: BUG #17231: ERROR: tuple concurrently updated
In-reply-to: <17231-f6234f3f4a9c84dd@postgresql.org>
References: <17231-f6234f3f4a9c84dd@postgresql.org>
Comments: In-reply-to PG Bug reporting form <noreply@postgresql.org>
	message dated "Fri, 15 Oct 2021 17:38:36 -0000"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <967137.1634322291.1@sss.pgh.pa.us>
Date: Fri, 15 Oct 2021 14:24:51 -0400
Message-ID: <967138.1634322291@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/967138.1634322291%40sss.pgh.pa.us>
Precedence: bulk

PG Bug reporting form <noreply@postgresql.org> writes:
> [ this: ]
>         c.execute('REFRESH MATERIALIZED VIEW CONCURRENTLY ab')
> [ concurrently with this: ]
>         c.execute('GRANT ALL ON ALL TABLES IN SCHEMA public TO web_admin')
> [ results in "tuple concurrently updated" ]

Yeah, that's not very surprising, because both commands need to update
ab's pg_class row.

We could only really prevent this by having GRANT take an exclusive
lock on the object(s) it's altering the ACLs of.  That cure would
likely be worse than the disease, especially for variants like
GRANT ON ALL TABLES, which'd need to ex-lock quite a lot of objects,
probably causing deadlock failures.  So I'm not sure that there's
any good way to improve this.

(There are, BTW, probably plenty of ways to cause this type of
failure when you do concurrent DDL commands on a single object.)

			regards, tom lane