From: cluster <skrald@amossen.dk>
Subject: Re: Query only slow on first run
Date: Tue, 27 Nov 2007 23:51:40 +0100
Organization: Hub.Org Networking Services
Lines: 27
Message-ID: <fii71p$oc1$1@news.hub.org>
References: <fihgt1$1mmh$1@news.hub.org>
	<20071127165543.GY11978@crankycanuck.ca>
	<23999.1196184368@sss.pgh.pa.us> <fihn7h$2jrn$1@news.hub.org>
	<056501c83128$55d58a90$8e00a8c0@tridecap.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
User-Agent: Thunderbird 2.0.0.6 (X11/20071022)
In-Reply-To: <056501c83128$55d58a90$8e00a8c0@tridecap.com>
To: pgsql-performance@postgresql.org

> As for optimizing the query, I noticed that all three joins are done by
> nested loops.  I wonder if another join method would be faster.  Have you
> analyzed all the tables?

Yes. I did a VACUUM FULL ANALYZE before running the test queries. Also I 
have just performed an ANALYZE just to be sure everything was really 
analyzed.

> You aren't disabling hash joins or merge joins are
> you?

Nope.

>  If you aren't, then as a test I would try disabling nested loops by
> doing "set enable_nestloop=false" and see if the query is any faster for
> you.

If I disable the nested loops, the query becomes *much* slower.

A thing that strikes me is the following. As you can see I have the 
constraint: q.status = 1. Only a small subset of the data set has this 
status. I have an index on q.status but for some reason this is not 
used. Instead the constraint are ensured with a "Filter: (q.status = 1)" 
in an index scan for the primary key in the "q" table. If the small 
subset having q.status = 1 could be isolated quickly using an index, I 
would expect the query to perform better. I just don't know why the 
planner doesn't use the index on q.status.