To: Chris <lists@deksai.com>
cc: pgsql-performance@postgresql.org
Subject: Re: stats collector suddenly causing lots of IO 
In-reply-to: <20100416151900.GA3179@chris-laptop.a2hosting.com> 
References: <20100413170105.GA29571@chris-laptop.a2hosting.com>
	<25952.1271370711@sss.pgh.pa.us>
	<20100416151900.GA3179@chris-laptop.a2hosting.com>
Comments: In-reply-to Chris <lists@deksai.com>
	message dated "Fri, 16 Apr 2010 11:19:13 -0400"
Date: Fri, 16 Apr 2010 11:29:47 -0400
Message-ID: <9857.1271431787@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Chris <lists@deksai.com> writes:
> After the file was made larger and I stopped the vacuum process, I started
> seeing the problem.  All other postgress processes were quiet, but the stats
> collector was constantly causing anywhere from 20-60 of the IO on the server.
> Since all the other postgres processes weren't really doing anything, and it is
> a busy web server which is predominately MySQL, I'm fairly curious as to what
> it is doing.

Yeah, the stats collector rewrites the stats file every half second, if
there have been any changes since last time --- so the bigger the file,
the more overhead.  (8.4 is smarter about this, but that doesn't help
you on 8.3.)

> I straced the stats collector process.  I wasn't sure what else to trace as
> there wasn't a single other postgres process doing anything.

That strace doesn't really prove much; it's what I'd expect.  Here's
what to do: start a PG session, and strace that session's backend *and*
the stats collector while you manually do VACUUM some-small-table.
The VACUUM command should try to send some messages to the stats collector
process.  I'm wondering if those get dropped somehow.

			regards, tom lane