To: pgsql-performance@postgresql.org
From: Mark Stosberg <mark@summersault.com>
Subject: Using statement_timeout as a performance tool?
Date: Thu, 01 Feb 2007 09:10:09 -0500
Lines: 27
Message-ID: <epssco$jes$1@sea.gmane.org>
Mime-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit
User-Agent: Thunderbird 1.5.0.8 (X11/20061109)


Hello,

I'm working on setting up replication with Slony, and will soon have a
slave that a lot of SELECT traffic will be sent to (over 500k/day).

The primary query we need to run is somewhat complex, but seems to
complete on average in well under a second.

However, every so often (less in 1 in 10,000 queries) we'll see the
query take 2 or 3 minutes.

It's not clear why this is happening-- perhaps there is something else
going on that is affecting this query.

I'm considering the use of  "statement_timeout" to limit the time of
this particular query, to suppress the rare "run away", and avoid tying
up the processor for that additional time.

I think it may be better to give up, quit spending cycles on it right
then, and return an "oops, try again in a few minutes" message instead.
From the data we have, seems like it has a strong chance of working again.

Is anyone else using "statement_timeout" as part of an overall
performance plan?

    Mark