[Mailman-Developers] CPU Usage

Barry A. Warsaw barry@zope.com
Sun, 7 Oct 2001 23:41:53 -0400

>>>>> "RP" == Rodolfo Pilas <rodolfo@linux.org.uy> writes:

    RP> Two mails about CPU usage:

    | Date: Tue, 25 Sep 2001 00:50:33 -0300
    | From: Rodolfo Pilas <rodolfo@linux.org.uy>
    | To: mailman-users@python.org
    | Subject: [Mailman-Users] CPU Usage in 2.1a2

    RP> Hello,

    RP> Perhaps somebody can explain me why I have a task (mailman) to
    RP> eat all of my CPU:

    RP> 60 processes: 56 sleeping, 4 running, 0 zombie, 0 stopped CPU
    RP> states: 5.1% user, 94.8% system, 0.0% nice, 0.0% idle Mem:
    RP> 259688K av, 154984K used, 104704K free, 0K shrd, 90484K buff
    RP> Swap: 385552K av, 12004K used, 373548K free 25160K cached

    RP> 25893 root 14 0 3688 3688 2468 R 99.0 1.4 2:42 python
    RP> bin/mailmanctl start 25899 root 2 0 924 924 732 R 0.7 0.3 0:01
Note, this is a Mailman 2.1 thing.

    RP> Sometimes I have two python task eating 50% of my CPU each
    RP> one.

    RP> It is normal?
    RP> How many time these task are overload the CPU?

    | Date: Mon, 24 Sep 2001 23:57:03 -0400
    | From: David Ball <dball@wcom.ca>
    | To: Rodolfo Pilas <rodolfo@linux.org.uy>
    | Subject: Re: [Mailman-Users] CPU Usage in 2.1a2

    DB> I have experienced the same problem recently (v2.0.6), and
    DB> ended up having to disable the Mailman web interface as
    DB> Python2.1 procs were taking down my machine (a mere P75 w/16MB
    DB> or ram, which may account for the problem).  Unless I killed
    DB> the processes immediately, all daemons would eventually shut
    DB> down (sshd, apache, even login), requiring me to reboot the
    DB> machine when I got home.

Mailman 2.1 and 2.0.6 use completely different qrunner systems, so
it's hard to understand how your two problems could be related.  I'm
not aware of any infloops in 2.0.6 and haven't seen any big problems
on {zope,python}.org.  I suppose the usual culprits like stale locks
and such could be at the heart of your problem.

OTOH, I haven't stress tested the 2.1 qrunner subsystem, so it's
possible there are problems there.  I'll believe I now have a test
framework where it might be possible to create very high loads under
control situations, so I should be able to uncover any performance
problems with 2.1's qrunner.