To: Andrew Sullivan <ajs@crankycanuck.ca>
Cc: pgsql-sql@postgresql.org
Subject: Re: row-attribute in EXPLAIN-output doesn't match count(*)
References: <200504141838.14705.andreak@officenet.no>
	<425EA496.2030501@archonet.com>
	<200504141921.38476.andreak@officenet.no>
	<20050414173358.GB9027@phlogiston.dyndns.org>
In-Reply-To: <20050414173358.GB9027@phlogiston.dyndns.org>
From: Greg Stark <gsstark@mit.edu>
Organization: The Emacs Conspiracy; member since 1992
Date: 14 Apr 2005 15:44:57 -0400
Message-ID: <8764ypf8hy.fsf@stark.xeocode.com>
Lines: 22
User-Agent: Gnus/5.09 (Gnus v5.9.0) Emacs/21.3
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii

Andrew Sullivan <ajs@crankycanuck.ca> writes:

> On Thu, Apr 14, 2005 at 07:21:38PM +0200, Andreas Joseph Krogh wrote:
> > 
> > So, what you're suggesting is that a restart of the webapp should make vacuum 
> > able to delete those dead rows?
> 
> Yes, but that'll only solve your problem for now.  You'll have the
> problem again soon.  What's keeping open the transaction?

This is presumably because of the long-standing issue that Postgres takes the
snapshot as soon as the BEGIN is issued. A lot of drivers issue a "COMMIT;
BEGIN;" right away even though it could be a long time before any actual work
is done.

Other databases (by which I mean Oracle) treat BEGIN as a noop. The snapshot
starts when the first SQL statement that needs a snapshot is executed. So
until a SELECT is issued the connection doesn't participate in any
transactional issues like keeping old versions of records around.

-- 
greg