[Zope-DB] Re: mysql concurrency

mkur at gazeta.pl mkur at gazeta.pl
Thu Jun 17 16:05:09 EDT 2004


Charlie Clark [charlie at egenix.com] wrote:
>
> Then you might want to look at increasing the number of threads in
Zope as
> this should help.

That was my first thought. Tried it. No improvement.

> Other than that read the source although the current RDBMS
> integration in Zope needs improving.

Right. Andy Dustman is doing something not too clear for me with his
python mysql API.
In ZMySQLDA 2.0.9 there is a huge shift in Zope TM / RDBS interaction.
Everything is wrapped in a classic MyISAM lock which seems to cause
really strange things to Zope TM. I had lots of errors in "events.log"
in even poorer concurrency. So I switched back to ZMySQLDA 2.0.8 with
mysqldb 0.9.2. This is a combination "always worked" for me but now I
learned it does not actually scale well. I think Solaris LWPs have
something to do with. I'm investigating it at the moment. I must say
that Zope otherwise flies on it. I have Plone apps on the same server.

> As part of the post Europython Zope
> sprint I did some work with Richard Waid on introducing support for
> positional parameters in ZSQL. We got 30% performance gains on
non-caching
> queries. Currently this is only available through a hack in Zope
core but we
> intend to be releasing a version of mxODBCZopeDA with mxZSQLs to take
> advantage of this. If you want I can explain what you need to hack
in Zope
> core.

I'm curious.

Thanks,

--
Michal Kurowski
<mkur at poczta.gazeta.pl>





More information about the Zope-DB mailing list