Message-ID: <44286A76.5010008@zeut.net>
Date: Mon, 27 Mar 2006 17:43:02 -0500
From: "Matthew T. O'Connor" <matthew@zeut.net>
User-Agent: Thunderbird 1.4 (Windows/20050908)
MIME-Version: 1.0
To: Mikael Carneholm <Mikael.Carneholm@WirelessCar.com>
CC: Postgresql Performance <pgsql-performance@postgresql.org>
Subject: Re: count(*) performance
References: <7F10D26ECFA1FB458B89C5B4B0D72C2B088295@sesrv12.wirelesscar.com>
In-Reply-To: <7F10D26ECFA1FB458B89C5B4B0D72C2B088295@sesrv12.wirelesscar.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Mikael Carneholm wrote:
> This is where a "last_vacuumed" (and "last_analyzed") column in
> pg_statistic(?) would come in handy. Each time vacuum or analyze has
> finished, update the row for the specific table that was
> vacuumed/analyzed with a timestamp in the last_vacuumed/last_analyzed
> column. No more guessing "maybe I haven't vacuumed/analyzed in a while",
> and each time a user complains about bad performance, one could request
> the user to do a "select s.last_vacuumed, s.last_analyzed from
> pg_statistic s, pg_attribute a, pg_class c where ..."
> 
> It SOUNDS easy to implement, but that has fooled me before... :-)


It is fairly easy to implement, however it has been discussed before and 
decided that it wasn't necessary.  What the system cares about is how 
long it's been since the last vacuum in terms of XIDs not time.  Storing 
a timestamp would make it more human readable, but I'm not sure the 
powers that be want to add two new columns to some system table to 
accommodate this.

Matt