From: Christopher Browne <cbbrowne@acm.org>
Subject: Re: Make COUNT(*) Faster?
Date: Fri, 08 Jul 2005 13:09:29 -0400
Organization: cbbrowne Computing Inc
Lines: 26
Message-ID: <m37jg1p6ly.fsf@mobile.int.cbbrowne.com>
References: <9EF2931D-1097-470D-8ECB-59D00166C96F@apple.com>
	<60vf3mklue.fsf@dba2.int.libertyrms.com>
	<42CE89CA.2070607@noao.edu> <20050708143508.GD22725@wolff.to>
	<42CE96AF.3020903@noao.edu>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
User-Agent: Gnus/5.1006 (Gnus v5.10.6) XEmacs/21.4 (Jumbo Shrimp,
	berkeley-unix)
Cancel-Lock: sha1:PhmFbpZKg3vUWbbyCcNPUd9HBmo=
To: pgsql-sql@postgresql.org

> I'm also claiming that a true count for any active table is
> meaningless and am *not* suggesting that effort be spent on trying
> to produce such a true count.

That's a pretty big assumption that would in fact be WRONG.

We have managers interested in counting the number of objects we have
around (As a domain registry, what objects would you imagine those
might be :-)), and they're keen on possibly even being able to
reconcile those counts from day to day based on transaction activity.

Leaping into some sort of vague guesstimation would destroy the
ability to do any kind of analysis of activity, and I daresay enrage
them.

There may be times that a really rough guess can suffice; there are
other times when exactness is absolutely vital.

Creating a "fast but WRONG COUNT(*)" which prevented getting the exact
answer that the present implementation provides would be a severe
misfeature.
-- 
output = reverse("gro.gultn" "@" "enworbbc")
http://linuxdatabases.info/info/rdbms.html
"The  test of a  principle  is whether it  applies  even to people you
don't like." -- Henry Spencer