Re: Drupal and PostgreSQL - performance issues?

On Wed, 15 Oct 2008, Gregory Stark wrote:

> Greg Smith <gsmith(at)gregsmith(dot)com> writes:
>> DB2 has automatically updated the "shmmax" kernel
>> parameter from "33554432" to the recommended value "268435456".
>
> This seems like a bogus thing for an application to do though.

It happens when you run a utility designed to figure out if the
application is compatible with your system and make corrections as it can
to make it work properly. If you want something like that to be easy to
use, the optimal approach to achieve that is to just barrel ahead and
notify the admin what you did.

> I'm pretty sure it would violate several Debian packaging rules.

You can wander to http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=67481
to see one of the many times someone has tried to get the SHMMAX situation
straightened out at the OS level. Quoth Martin Pitt: "Debian
packages...must not alter kernel parameters at will; if they did, they
would destroy each others settings.", followed by the standard wontfix for
actually changing anything. They did at least improve the error reporting
when the server won't start because of this problem there.

> Generally it seems crazy for a distribution to ship configured one way
> by default but have packages change that behind the user's back. What if
> the admin set SHMMAX that way because he wanted it? What happens when a
> new distribution package has a new default but doesn't adjust it because
> it sees the "admin" has changed it -- even though it was actually
> Postgres which made the change?

If there were ever any Linux distributions that increased this value from
the tiny default, you might have a defensible position here (maybe
Oracle's RHEL fork does, they might do something here). I've certainly
never seen anything besides Solaris ship with a sensible SHMMAX setting
for database use on 2008 hardware out of the box. It's really quite odd,
but as numerous probes in this area (from the above in 2000 to Peter's
recent Linux bugzilla jaunt) show the resistance to making the OS default
to any higher is considerable.

--
* Greg Smith gsmith(at)gregsmith(dot)com http://www.gregsmith.com Baltimore, MD