[Mailman-Users] Call for suggestions

[Marc MERLIN]

On Tue, May 08, 2001 at 04:32:56PM -0600, Ashley M. Kirchner wrote:
> 
>     I want to setup multiple servers running (the same) mailman lists.
> The ways I can think of doing this is either:
 
Been there, done that, didn't work.
 
>     Option 1:
>        Setup a master machine with everything on it and export the
>        mailman structure for NFS so the other machines can mount it,
>        and use it.
> 
>        Problem: What happens if the NFS connection dies?  Msgs get
>                 lost.  That's not desirable.

They don't get lost, connections just hang, you take the server off the
rotation, it's not that bad.
NFS didn't work though because under high load, while each server had its
own queue directory and was running its own qrunner, mailman processes had
concurent access to the list config.db files (which have to be shared).
Even though the mailman locking is well done, it hit races in NFS.

See:
http://mail.python.org/pipermail/mailman-users/2000-November/007618.html
(read the thread)
http://mail.python.org/pipermail/mailman-users/2000-November/007942.html
http://mail.python.org/pipermail/mailman-developers/2000-December/003383.html
(read thread)


>     Option 2:
>        Have the remote machines mirror the master (through [s]ftp or
>        [s]rcp).  This will avoid having to deal with NFS dropping.
> 
>        Problem: The files won't be as up-to-date as the master, unless
>                 one starts mirroring continuously, which would be
>                 ridiculous.
 
You will have problems with config.db files not being up to date.

The only  thing that's actually  really slow  in mailman is  qrunner because
it's single threaded, and most of  the problems come from the HTML archiving
in pipermail.
If you disable  HTML archiving, or archiving altogether, even  if qrunner is
single threaded,  you can deliver lots  of Email with mailman  (about half a
million per day accross 10,000 lists for SF) on a single machine.
 
I haven't given up on the shared spool over NFS though for failover reasons
more than performance.

Marc
-- 
Microsoft is to operating systems & security ....
                                      .... what McDonalds is to gourmet cooking
  
Home page: http://marc.merlins.org/   |   Finger marc_f at merlins.org for PGP key