Message-ID: <4C8EB813.6070204@agliodbs.com>
Date: Mon, 13 Sep 2010 16:47:31 -0700
From: Josh Berkus <josh@agliodbs.com>
User-Agent: Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10.5; en-US;
	rv:1.9.1b3pre) Gecko/20090223 Thunderbird/3.0b2
MIME-Version: 1.0
To: pgsql-performance@postgresql.org
Subject: Re: Where does data in pg_stat_user_tables come from?
References: <4C8EAE5E.70201@agliodbs.com>
	<1284421288.28610.501.camel@jd-desktop.unknown.charter.com>
In-Reply-To: <1284421288.28610.501.camel@jd-desktop.unknown.charter.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit

On 9/13/10 4:41 PM, Joshua D. Drake wrote:
> On Mon, 2010-09-13 at 16:06 -0700, Josh Berkus wrote:
>> All,
>>
>> I've been looking at pg_stat_user_tables (in 8.3, because of a project I
>> have), and it appears that autovacuum, and only autovaccum, updates the
>> data for this view.  This means that one can never have data in
>> pg_stat_user_tables which is completely up-to-date, and if autovacuum is
>> off, the view is useless.
> 
> As I recall its kept in shared_buffers (in some kind of counter) and
> updated only when it is requested or when autovacuum fires. This was
> done because we used to write stats every 500ms and it was a bottleneck.
> (IIRC)

Yes, looks like it only gets updated on SELECT or on autovacuum.

Thing is, a full VACUUM ANALYZE on the database, or even just ANALYZE,
should update some of the counters.  And currently it doesnt, resulting
in pg_class.reltuples often being far more up to date than
pg_stat_user_tables.n_live_tup.  And frankly, no way to reconcile those
two stats.

-- 
                                  -- Josh Berkus
                                     PostgreSQL Experts Inc.
                                     http://www.pgexperts.com