To: josh@agliodbs.com
Cc: "Richard Sydney-Smith" <richard@ibisaustralia.com>,
	<pgsql-sql@postgresql.org>
Subject: Re: Slow sub-selects, max and count(*)
References: <001001c3eb8a$255d6dc0$5801a8c0@athlon2000>
	<200402042113.01369.josh@agliodbs.com>
In-Reply-To: <200402042113.01369.josh@agliodbs.com>
From: Greg Stark <gsstark@mit.edu>
Organization: The Emacs Conspiracy; member since 1992
Date: 28 Mar 2004 01:31:31 -0500
Message-ID: <87brmhtrcs.fsf@stark.xeocode.com>
Lines: 23
User-Agent: Gnus/5.09 (Gnus v5.9.0) Emacs/21.3
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii


Josh Berkus <josh@agliodbs.com> writes:

> Max() and Count() cannot use indexes for technical reasons.   Browse through 
> the archives of SQL, PERFORM, and HACKERS for about 12,000 discussions on the 
> subject.

Please don't confuse the issue by throwing Max() and Count() into the same
basket.

The issue with Min()/Max() is that the result could be generated efficiently
by scanning indexes but it's just hard, especially given generic aggregate
functions, and the work simply hasn't been done, or even started, yet.

The issue with Count() is that people want the result to be cached in a single
per-table counter, but that can't be done as simply as that because of
transactions. People have discussed complex solutions to this but it's a much
more complex problem than it appears.

They're really two entirely separate issues.

-- 
greg