[Mailman-Users] Bounce Processing Problems

[Lorenz Adena]

On Thu 2005-09-01 (18:28), Brad Knowles wrote:
> 	Try moving the current "bounce" directory to something like 
> "bounce.old", and create a new one with the same ownership and 
> permissions, then stop and restart Mailman.  I'd be willing to bet 
> that directory lock contention has been a *huge* part of your problem 
> -- I certainly believe that it was for us on python.org.

since I had a similar problem a few weeks ago I'll try to remember what
i found out then.

- Incoming bounces are stored in mailman/qfiles/bounces
- BounceRunner empties this directory and collects all the bounces
  in mailman/data/bounce-events-<pid>.pck where the real processing
  starts. A big number of bounces in a short time results to a
  correspondingly big bounce-events-file. This step is working pretty
  fast, so the bounce-events-file can grow really big in a short time.
- Bounce Events are now sorted by list and pending Requests are stored in
  mailman/lists/<listname>/pending.pck
  If this file grows beyond a certain limit the BounceRunner spends most
  of his time in Reading (and writing?) this file. I'm not sure where
  the exact limit is, but I suspect something above a few MB - I
  encountered files as big as 50MB.

If the BounceRunner is in the described state, the following problems
occur:

- new bounces are processed from mailman/qfiles/bounces at a very
  slow rate of one bounce a minute or even slower
- the admin and list-pages of the affected list(s) on the web-server
  produce a timeout - all other web-pages work fine
- shutting down mailman does not work exactly like intended, since
  the BounceRunner is not killed immediately but finishes whatever
  he is doing, which can take several hours. kill -9 should work but
  will result in missing bounces.
- the rest of the Mailman programs and lists work fine

The suggested move of the qfiles/bounce directory to bounce.old worked
for me, but it may take some time for the BounceRunner to finish. You
might even move the files from bounce.old to the current bounce
directory to make sure they are processed, but you have to make sure the
pending.pck is removed from the list-directory before or you will run
into the same problems again.

Hope this information helps in solving the problem.

Regards

Lorenz
-- 
Lorenz Adena                             Information & Medien
Rechenzentrum
Universität Stuttgart               Tel:  (++49-711) 685-5993
Allmandring 30                      Fax:  (++49-711) 68702041
70550 Stuttgart

*** RUS-Mailadresse: poststelle at rus.uni-stuttgart.de ***