Subject: Re: controlling autovacuum during the day.
From: Brad Nicholson <bnichols@ca.afilias.info>
To: Alvaro Herrera <alvherre@commandprompt.com>
Cc: Michael Fuhr <mike@fuhr.org>, Tom Lane <tgl@sss.pgh.pa.us>,
	John Lister <john.lister-ps@kickstone.com>, pgsql-admin@postgresql.org
In-Reply-To: <20081217163130.GF4453@alvh.no-ip.org>
References: <2BB7361E57C4417A975C0474610BA4C3@squarepi.com>
	<27759.1229521625@sss.pgh.pa.us>
	<742A9D77A1BA4171AF52BBBF6F5F2C75@squarepi.com>
	<28510.1229525243@sss.pgh.pa.us>
	<20081217152903.GA37436@winnie.fuhr.org>
	<20081217163130.GF4453@alvh.no-ip.org>
Content-Type: text/plain; charset=utf-8
Date: Wed, 17 Dec 2008 14:35:51 -0500
Message-Id: <1229542551.6302.13.camel@bnicholson-desktop>
Mime-Version: 1.0
Content-Transfer-Encoding: quoted-printable

On Wed, 2008-12-17 at 13:31 -0300, Alvaro Herrera wrote:
> Michael Fuhr wrote:
> > On Wed, Dec 17, 2008 at 09:47:23AM -0500, Tom Lane wrote:
> > > "John Lister" <john.lister-ps@kickstone.com> writes:
> > > > Cheers for the quick reply. I've tweaked them quite a bit, but we h=
ave quite=20
> > > > a few heavily updated tables that i'd like vacuuming to keep them i=
n check.=20
> > > > Unfortunately the autovacuum does a FULL vacuum every so often lock=
ing the=20
> > > > tables for quite a long time, i'd like to move these to the evening=
 if=20
> > > > possible.
> > >=20
> > > Huh?  Autovacuum *never* does VACUUM FULL.
> >=20
> > Perhaps autovacuum is shrinking the table after finding lots of empty
> > pages at the end, as in what VACUUM VERBOSE is logging here:
> >=20
> > INFO:  "foo": truncated 11944384 to 8877366 pages
> >=20
> > I think this acquires an AccessExclusiveLock.  I've seen this take
> > hours in the case of a table with a lot of empty pages.
>=20
> There was a bug in earlier versions which caused lazy vacuum to do
> cost-based delays during this phase, which caused the lock to be held
> for ridiculous lengths of time.  This was fixed in 8.2.something; it
> shouldn't sleep anymore, but it does need to scan those pages in order
> to truncate, so it can still take a while.

Does this mean that if you are using autovac on 8.1, you should not use
the cost delay feature?
--=20
=EF=BB=BFBrad Nicholson  416-673-4106
Database Administrator, Afilias Canada Corp.