Date: Fri, 20 Sep 2024 14:50:30 -0700
From: Noah Misch <noah@leadboat.com>
To: Nathan Bossart <nathandbossart@gmail.com>
Cc: vaibhave postgres <postgresvaibhave@gmail.com>,
	pgsql-bugs@lists.postgresql.org, vsekar@microsoft.com
Subject: Re: vacuumdb: permission denied for schema "pg_temp_7"
Message-ID: <20240920215030.8d.nmisch@google.com>
References: 
 <CAM_eQjwfAR=y3G1fGyS1U9FTmc+FyJm9amNfY2QCZBnDDbNPZg@mail.gmail.com>
 <20240920190731.37.nmisch@google.com>
 <Zu3iMzfiGBTbg3iy@nathan>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <Zu3iMzfiGBTbg3iy@nathan>
User-Agent: Mutt/2.2.12 (2023-09-09)
Archived-At: 
 <https://www.postgresql.org/message-id/20240920215030.8d.nmisch%40google.com>
Precedence: bulk

On Fri, Sep 20, 2024 at 03:59:32PM -0500, Nathan Bossart wrote:
> On Fri, Sep 20, 2024 at 12:07:31PM -0700, Noah Misch wrote:
> > v12 introduced this regression.  I suspect it started when commit e0c2933 "Use
> > catalog query to discover tables to process in vacuumdb" switched vacuumdb
> > from a simple "VACUUM;" command to per-table commands.  Non-superuser vacuumdb
> > must be rare indeed for this to go unnoticed long enough to leave all
> > supported branches affected.
> 
> I think the bug actually predates that commit, but it was only broken when
> --jobs > 1.  Commit e0c2933 just broke the --jobs == 1 case, too.

Agreed.

> > That helps, but we'd probably want to do something more general about vacuumdb
> > and schema USAGE permission.
> 
> Hm.  I think filtering out schemas for which you lack USAGE makes sense
> when neither --schema nor --table are specified, but if the user lists an
> object they can't vacuum, we should probably fail.  My current thinking is
> that we could still filter when --exclude-schema is used, but I'm curious
> what others think.

That all sounds good to me.

> You might also be interested in this thread about VACUUM and USAGE [0].

> [0] https://postgr.es/m/flat/56596b81-088f-4c0c-9a88-b5f27a7a62e9%40oss.nttdata.com

The outcome is odd, but I'm not worried about it.