From: Dimitri Fontaine <dfontaine@hi-media.com>
To: Josh Berkus <josh@agliodbs.com>
Cc: postgres performance list <pgsql-performance@postgresql.org>
Subject: Re: PostgreSQL as a local in-memory cache
Organization: Hi-Media
References: <cb0fb58c-9134-4314-a1d0-08fc39f911a6@40g2000pry.googlegroups.com>
	<18287446-8677-46d6-a581-c4777f60ebda@40g2000pry.googlegroups.com>
	<4C1A5B81.8050908@agliodbs.com>
User-Mail-Address: dfontaine@hi-media.com
Date: Thu, 17 Jun 2010 21:01:19 +0200
In-Reply-To: <4C1A5B81.8050908@agliodbs.com> (Josh Berkus's message of "Thu,
	17 Jun 2010 10:29:37 -0700")
Message-ID: <m2typ18ow0.fsf@hi-media.com>
User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/23.1.90 (darwin)
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable

Hi,

Josh Berkus <josh@agliodbs.com> writes:
> a) Eliminate WAL logging entirely
> b) Eliminate checkpointing
> c) Turn off the background writer
> d) Have PostgreSQL refuse to restart after a crash and instead call an
> exteral script (for reprovisioning)

Well I guess I'd prefer a per-transaction setting, allowing to bypass
WAL logging and checkpointing. Forcing the backend to care itself for
writing the data I'm not sure is a good thing, but if you say so.

Then you could have the GUC set for a whole cluster, only a database
etc. We already have synchronous_commit to trade durability against
performances, we could maybe support protect_data =3D off too.

The d) point I'm not sure still applies if you have per transaction
setting, which I think makes the most sense. The data you choose not to
protect is missing at restart, just add some way to register a hook
there. We already have one (shared_preload_libraries) but it requires
coding in C.=20

Calling a user function at the end of recovery and before accepting
connection would be good I think. A user function (per database) is
better than a script because if you want to run it before accepting
connections and still cause changes in the database=E2=80=A6

Regards,
--=20
dim