Date: Thu, 28 Jun 2007 09:52:54 -0400 (EDT)
From: Greg Smith <gsmith@gregsmith.com>
To: pgsql-performance@postgresql.org
Subject: Re: PostgreSQL 8.0 occasionally slow down
In-Reply-To: <97e6e20f0706280101y543b871ao4cbf85e57f07f7e9@mail.gmail.com>
Message-ID: <Pine.GSO.4.64.0706280942320.6275@westnet.com>
References: <97e6e20f0706272354m5e42c91em177c8b80abb2ea52@mail.gmail.com>
	<468361A9.1040308@archonet.com>
	<97e6e20f0706280101y543b871ao4cbf85e57f07f7e9@mail.gmail.com>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII; format=flowed

On Thu, 28 Jun 2007, Ho Fat Tsang wrote:

> I have tuned the checkpoint_timeout to 30 second which is ten times less 
> than default and the issue is still reproduced.

Doing a checkpoint every 30 seconds is crazy; no wonder your system is 
pausing so much.  Put the timeout back to the default.  What you should do 
here is edit your config file and set checkpoint_warning to its maximum of 
3600.  After that, take a look at the log files; you'll then get a warning 
message every time a checkpoint happens.  If those line up with when 
you're getting the slowdowns, then at least you'll have narrowed the cause 
of your problem, and you can get some advice here on how to make the 
overhead of checkpoints less painful.

The hint it will give is probably the first thing to try: increase 
checkpoint_segments from the default to something much larger (if it's at 
3 now, try 10 instead to start), and see if the problem stops happening as 
frequently.  Your problem looks exactly like a pause at every checkpoint, 
and I'm not sure what Richard was thinking when he suggested having them 
more often would improve things.

--
* Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD