To: dracula007@atlas.cz
Cc: pgsql-sql@postgresql.org
Subject: Re: **SPAM** Faster count(*)? 
In-reply-to: <196456155.20050810012958@karneval.cz> 
References: <000101c59d33$36a4a560$9b00015a@osl.com>
	<196456155.20050810012958@karneval.cz>
Comments: In-reply-to dracula007@atlas.cz
	message dated "Wed, 10 Aug 2005 01:29:58 +0200"
Date: Tue, 09 Aug 2005 22:49:14 -0400
Message-ID: <21626.1123642154@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

dracula007@atlas.cz writes:
> I believe running count(*) means fulltable scan, and there's no way
> to do it without it. But what about some "intermediate" table, with
> the necessary counts?

There's a fairly complete discussion in the PG list archives of a
reasonably-efficient scheme for maintaining such counts via triggers.
It wasn't efficient enough that we were willing to impose the overhead
on every application ... but if you really NEED a fast count(*) you
could implement it.  I'd like to see someone actually do it and put
up working code on pgfoundry; AFAIK it's only a paper design so far.

If you only want a very-approximate count, the best bet is to rely on
the planner's estimates, eg

regression=# explain select * from tenk1;
                         QUERY PLAN                          
-------------------------------------------------------------
 Seq Scan on tenk1  (cost=0.00..458.00 rows=10000 width=244)
                                            ^^^^^

Current best practice is to run the explain and parse out the "rows"
figure using a perl (or axe-of-choice) regexp, though we could be
persuaded to supply a simpler API if there's enough demand for it.

			regards, tom lane