Date: Fri, 27 Oct 2006 23:19:20 +0100
From: Gavin Hamill <gdh@laterooms.com>
To: pgsql-performance@postgresql.org
Subject: Re: VACUUMs take twice as long across all nodes
Message-Id: <20061027231920.b806d071.gdh@laterooms.com>
In-Reply-To: <7564.1161972463@sss.pgh.pa.us>
References: <20061026095827.9e85e1c0.gdh@laterooms.com>
	<20196.1161874041@sss.pgh.pa.us>
	<20061026160609.5e94e581.gdh@laterooms.com>
	<20061026191728.GD26892@nasby.net>
	<20061026213556.174bd7b5.gdh@laterooms.com>
	<20061026220937.GA6866@phlogiston.dyndns.org>
	<20061027102025.8183c5d9.gdh@laterooms.com>
	<7564.1161972463@sss.pgh.pa.us>
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit

On Fri, 27 Oct 2006 14:07:43 -0400
Tom Lane <tgl@sss.pgh.pa.us> wrote:

> So the time is all in index vacuuming, eh?  I think what's happening
> is that the physical order of the index is degrading over time, and
> so the vacuum scan takes longer due to more seeking.  Can you afford
> to do a REINDEX?  If this theory is correct that should drive the
> time back down.

Tom,

You wonderful, wonderful man.

I tried a test reindex on "Allocation", and noticed a vacuum had
turbo-charged... then reindexed the whole db, did a vacuum, and lo! The
whole db had turbo-charged :)

When I say 'turbo-charged', I mean it. The vacuum times have dropped to
20% of what we were seeing even before it 'got much slower a
couple of days ago.'

It sucks that the new reindex code is only in 8.2, but now that I know
this is an issue in 8.1 I can plan for it.

Thanks so much :)

Cheers,
Gavin.