From: Josh Berkus <josh@agliodbs.com>
Organization: PostgreSQL @ Sun
To: Ron Mayer <rm_pg@cheapcomplexdevices.com>
Subject: Re: Overhauling GUCS
Date: Tue, 10 Jun 2008 09:33:41 -0700
User-Agent: KMail/1.8.2
Cc: "Joshua D. Drake" <jd@commandprompt.com>, Tom Lane <tgl@sss.pgh.pa.us>,
	Peter Eisentraut <peter_e@gmx.net>, pgsql-hackers@postgresql.org,
	Greg Smith <gsmith@gregsmith.com>,
	Andreas Pflug <pgadmin@pse-consulting.de>, "Decibel!" <decibel@decibel.org>
References: <48409D1E.3070208@agliodbs.com>
	<4849598F.9080800@commandprompt.com>
	<4849A6F3.8010800@cheapcomplexdevices.com>
In-Reply-To: <4849A6F3.8010800@cheapcomplexdevices.com>
MIME-Version: 1.0
Content-Type: text/plain;
  charset="iso-8859-1"
Content-Transfer-Encoding: 7bit
Content-Disposition: inline
Message-Id: <200806100933.42232.josh@agliodbs.com>

Ron,

> I wonder if the fastest way to generate the configurator
> would be to simply ask everyone to post their tuned
> postgresql.conf files along with a brief description of
> the use case for that file.  The we could group the
> use-cases into various classes; and average the values
> of the submitted files.  Then the configurator's one
> question "choose which use case most closely matches
> yours from this list".

This has been proposed a number of times, but is still infeasable for the 
simple reason that a lot of settings are dependant on how much RAM you have 
on the machine.  "Sample configs" would just result in people trying to set 
their shared_buffers to 150% of system RAM if they have a lightweight system, 
or in having 500 different sample configurations, which would produce even 
*more* user confusion.

A generator makes more sense.

-- 
Josh Berkus
PostgreSQL @ Sun
San Francisco