From: =?utf-8?q?C=C3=A9dric_Villemain?= <cedric@2ndquadrant.com>
Reply-To: cedric@2ndquadrant.com
Organization: 2ndQuadrant
To: pgsql-performance@postgresql.org
Subject: Re: random_page_cost = 2.0 on Heroku Postgres
Date: Fri, 10 Feb 2012 20:48:07 +0100
User-Agent: KMail/1.13.7 (Linux/3.1.0-1-amd64; KDE/4.6.5; x86_64; ; )
Cc: Josh Berkus <josh@agliodbs.com>, Peter van Hardenberg <pvh@pvh.ca>,
	Jeff Janes <jeff.janes@gmail.com>, Scott Marlowe <scott.marlowe@gmail.com>
References: 
 <CAAcg=kWsjAXRNt0sSjsF-Ft+NfZi30Owa=-PGvghSJ6A0jeVYQ@mail.gmail.com>
	<CAAcg=kXOiAi66tzP6hdFTSECjvKjLx65jUcMYQ1g3j4XdzOWDw@mail.gmail.com>
	<4F3570E2.2070301@agliodbs.com>
In-Reply-To: <4F3570E2.2070301@agliodbs.com>
MIME-Version: 1.0
Content-Type: Text/Plain;
  charset="utf-8"
Content-Transfer-Encoding: quoted-printable
Message-Id: <201202102048.07495.cedric@2ndquadrant.com>

Le vendredi 10 f=C3=A9vrier 2012 20:32:50, Josh Berkus a =C3=A9crit :
> On 2/9/12 2:41 PM, Peter van Hardenberg wrote:
> > Hmm, perhaps we could usefully aggregate auto_explain output.
>=20
> The other option is to take a statistical approach.  After all, what you
> want to do is optimize average response times across all your user's
> databases, not optimize for a few specific queries.
>=20
> So one thought would be to add in pg_stat_statements to your platform
> ... something I'd like to see Heroku do anyway.  Then you can sample
> this across dozens (or hundreds) of user databases, each with RPC set to
> a slightly different level, and aggregate it into a heat map.
>=20
> That's the way I'd do it, anyway.

in such set up, I sometime build a ratio between transactions processed and=
=20
CPU usage, many indicators exists, inside and outside DB, that are useful t=
o=20
combine and use just as a 'this is normal behavior'. It turns to be easy in=
=20
the long term to see if things go better or worse.

=2D-=20
C=C3=A9dric Villemain +33 (0)6 20 30 22 52
http://2ndQuadrant.fr/
PostgreSQL: Support 24x7 - D=C3=A9veloppement, Expertise et Formation