Message-ID: <4DCAB16F.2060300@peak6.com>
Date: Wed, 11 May 2011 10:55:27 -0500
From: Shaun Thomas <sthomas@peak6.com>
Reply-To: <sthomas@peak6.com>
Organization: Peak6
User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US;
	rv:1.9.2.7) Gecko/20100720 Lightning/1.0b2 Thunderbird/3.1.1
MIME-Version: 1.0
To: "Prodan, Andrei" <Andrei.Prodan@awinta.com>
CC: <pgsql-performance@postgresql.org>
Subject: Re: 'Interesting' prepared statement slowdown on large
 table join
References: 
 <D33393149C01874DB3D72FE46A1B268B0C784D@vpmail05-x.intra.ads-root.de>
In-Reply-To: 
 <D33393149C01874DB3D72FE46A1B268B0C784D@vpmail05-x.intra.ads-root.de>
Content-Type: text/plain; charset="ISO-8859-1"; format=flowed
content-transfer-encoding: 7bit

On 05/11/2011 06:08 AM, Prodan, Andrei wrote:

> Index Scan using attr_name_value on big_table  (cost=0.00..22.85
> rows=4 width=7) (actual time=0.176..757.646 rows=914786 loops=1)

Holy inaccurate statistics, Batman!

Try increasing your statistics target for attr_name and attr_value in 
your big table. I know you said you set it to 300 on party_id, but what 
happened here is that the optimizer thought this particular name/value 
combo in your big table would return less rows, and it was horribly, 
horribly wrong.

You might think about bumping up your default_statistics_target anyway 
to prevent problems like this in general. But definitely increase it on 
those two columns and reanalyze. My guess is that your big_table is big 
enough that each analyze gets a different random sample of the various 
attr_name and attr_value combinations, so occasionally it'll get too few 
and start badly skewing query plans.

-- 
Shaun Thomas
OptionsHouse | 141 W. Jackson Blvd. | Suite 800 | Chicago IL, 60604
312-676-8870
sthomas@peak6.com

______________________________________________

See  http://www.peak6.com/email_disclaimer.php
for terms and conditions related to this email