Date: Fri, 14 Dec 2018 23:06:32 -0300
From: Alvaro Herrera <alvherre@2ndquadrant.com>
To: Robert Haas <robertmhaas@gmail.com>
Cc: Tom Lane <tgl@sss.pgh.pa.us>,
	PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Subject: Re: removal of dangling temp tables
Message-ID: <20181215020632.5vlffyvfpn4k42cn@alvherre.pgsql>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: 
 <CA+TgmoYuAupG1+aovKuauNZX27GGCda5BSfT=tXtFrkd+8HmaQ@mail.gmail.com>
User-Agent: NeoMutt/20180716
Precedence: bulk

On 2018-Dec-14, Robert Haas wrote:

> On Fri, Dec 14, 2018 at 6:35 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> > Hm.  It *could*, if we wanted it to run some transactions after
> > finishing recovery.
> 
> It'd have to launch a separate process per database.  That would be
> useful infrastructure for other things, too, like automatic catalog
> upgrades in minor releases, but I'm not volunteering to write that
> infrastructure right now.

This looks like a major project.

> > Alternatively, maybe we could have backends flag whether they've
> > taken ownership of their temp schemas or not, and let autovacuum
> > flush old temp tables if not?
> 
> Yes, that seems like a possibly promising approach.

I did propose in my OP the idea of a PGPROC boolean flag that indicates
whether the temp namespace has been set up.  If not, have autovac remove
those tables.  I like this option better, but I fear it adds more
ProcArrayLock contention.  Maybe I can just use a new LWLock to
coordinate that particular member of the ProcGlobal array ... (but then
it can no longer be a boolean.)

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