DomainKey-Signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma;
	h=mime-version:in-reply-to:references:date:message-id:subject:from:to
	:cc:content-type:content-transfer-encoding;
	b=Qg0+1/6rVlR3wToMdrXExUANEj8x6kwWSJvIjIS71dJ8mLzYeaPo/qFVLH1ireyws3
	jqZHDHsyleeEud6rECXciIQmaorqLJqbgck8f/vl6QLzAxXpVYw9yHvEAYZity3GIdRj
	vsEOxyqk3iErWkMm27t7juFM+aYkxkSsJbves=
MIME-Version: 1.0
In-Reply-To: <49E42C72.9070902@obviously.com>
References: <49E42C72.9070902@obviously.com>
Date: Tue, 14 Apr 2009 01:29:02 -0600
Message-ID: <dcc563d10904140029j78b70370hf2b385a4fad82e76@mail.gmail.com>
Subject: Re: Postgres entering zombie state once a week in production
	evnvironment
From: Scott Marlowe <scott.marlowe@gmail.com>
To: Bryce Nesbitt <bryce2@obviously.com>
Cc: pgsql-sql@postgresql.org
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable

On Tue, Apr 14, 2009 at 12:25 AM, Bryce Nesbitt <bryce2@obviously.com> wrot=
e:
> We have a medium scale installation of Postgres 8.3 that is freezing abou=
t
> once a week. =A0I'm looking for any hints on how to diagnose the situatio=
n, as
> nothing is logged.
>
> The system is matched pair of Sunfire servers, running Debian Etch with a
> 2.6.18-6-amd64 kernel, PostgreSQL 8.3.4, and DRBD 8.0.13.
>
> During a failed state, pg_stat_activity will show hundreds of statements
> pending. =A0query_start will show the statements arriving at a normal rat=
e (a
> few per second), but clearly they never complete. =A0The bulk of these
> statement are a simple select that starts each web session, a statement t=
hat
> generally completes in tenths of milliseconds. =A0Restarting postgres res=
tores
> normal operation, at the loss of all chance of figuring out what was wron=
g.

What does pg_locks say during this time?  Specifically about locks
that aren't granted?