[Zope] Performance under load using SQL (Postgresql)

[Joel Burton]

On Tue, 27 Mar 2001, Mayers, Philip J wrote:

> I seem to be able to kill Zope *very* easily (with about 5 requests/sec) if
> that request does a lot of SQL activity. Now, this activity is not
> avoidable, but that load is pretty puny.
> 
> Also, it doesn't just seem to die either - it seems to hang completely under
> some circumstances.
> 
> I'm using Postgresql (makes no different whether it's 7.0 or the 7.1RC1)
> with ZPoPyDA and/or psycopg - they both suffer from the problem. What
> happens is that the child threads of ZServer die, leaving me with the
> top-level process (doing a waitpid on it's child) and the child (doing a
> select on the PCGI socket)
> 
> How can I debug this? The machine is plenty fast enough to handle this load,
> and with static pages it's fine - just the ones that make SQL calls. Surely
> it should degrade gracefully, rather than just *dying*?

I'm not running anything *too* heavy, but have seen our our
PostgreSQL-backed Zope site hit more than 5req/sec during testing w/o a
problem. W/a python test script, we sent about 15 requests per second w/o
problem. (We were testing for stability for some of the more complex
pages; our site dreams it ever got 15req per second.)

PG7.1beta4 (at time, I believe. We're at 7.1RC1 now)
ZPoPy, more recent version
Zope 2.3.1b2 (at the time of testing)

Running on Linux, on a KIII/550 128MB, otherwise unoccupied machine.

Not sure if we were lucky or, err, if you're not.

HTH,
-- 
Joel Burton   <jburton@scw.org>
Director of Information Systems, Support Center of Washington