Message-ID: <441542f15012666c621814327415a2cbb0960e10.camel@j-davis.com>
Subject: Re: pgsql: Recalculate search_path after ALTER ROLE.
From: Jeff Davis <pgsql@j-davis.com>
To: Jeff Davis <jdavis@postgresql.org>, pgsql-committers@lists.postgresql.org
Date: Wed, 09 Aug 2023 23:08:10 -0700
In-Reply-To: <d4bc331366068e9605d65bfe471687b4d32528d8.camel@j-davis.com>
References: <E1qTpps-001FmA-IW@gemulon.postgresql.org>
	 <d4bc331366068e9605d65bfe471687b4d32528d8.camel@j-davis.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.44.4-0ubuntu2 
MIME-Version: 1.0
Archived-At: 
 <https://www.postgresql.org/message-id/441542f15012666c621814327415a2cbb0960e10.camel%40j-davis.com>
Precedence: bulk

On Wed, 2023-08-09 at 16:12 -0700, Jeff Davis wrote:
> I'm not sure yet, but this looks like a possible pre-existing bug, or
> some incorrect assumption I made in fa2e874946.

Before my commit the behavior is the same. Here's a minimal repro
(debug_parallel_query set to 'on' or 'regress'):

  s1:
    CREATE USER u1;
    SET SESSION AUTHORIZATION u1;

  s2:
    ALTER ROLE u1 RENAME TO u2;

  s1:
    SELECT 1;
    ERROR:  role "u1" does not exist
    CONTEXT:  while setting parameter "session_authorization" to "u1"

Given that it works fine without parallel query, I'm inclined to call
this a parallel query bug. The error is coming from RestoreGUCState().

That being said, the buildfarm is red after my commit, so unless
someone sees a quick fix here, perhaps I should remove my isolation
test? I don't want to revert my whole commit, because it seems correct
and fixes a separate bug.

Regards,
	Jeff Davis