Date: Tue, 2 Apr 2019 14:00:51 -0300
From: Alvaro Herrera <alvherre@2ndquadrant.com>
To: Andres Freund <andres@anarazel.de>
Cc: Tom Lane <tgl@sss.pgh.pa.us>, r.zharkov@postgrespro.ru,
	pgsql-bugs@lists.postgresql.org
Subject: Re: BUG #15727: PANIC: cannot abort transaction 295144144, it was
 already committed
Message-ID: <20190402170051.GA22324@alvherre.pgsql>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <20190402161320.o5wkbp7evuzqe4c2@alap3.anarazel.de>
User-Agent: Mutt/1.9.4 (2018-02-28)
Precedence: bulk

On 2019-Apr-02, Andres Freund wrote:

> On 2019-04-02 12:51:08 -0300, Alvaro Herrera wrote:

> > AFAICS this error can only come from ExecDelete(), because the value 1
> > is TM_Invisible and the other callsites where the "unexpected
> > table_lock_tuple" error appears use different wording for that one.
> 
> Hm? Why couldn't it be the ExecUpdate() case?

You're right, that one too.

> > Maybe it's the result of a deferred constraint being checked at that
> > time ... maybe it's trying to honor an "on cascade delete" setting for
> > an FK, and the affected tuple has already been updated or deleted?
> 
> Then it ought to get TM_Deleted, no? We ought to wait till that
> transaction commits, and then roll back.

I was thinking that this would have happened in the same transaction;
but yeah I didn't spend too much time analyzing the exact code flow.
Anyway I agree that there's something odd going on, and perhaps you just
unmasked an earlier bug.

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services