[Mailman-Users] Mailman 2.1.6 slowness...?

[Jeff Squyres]

Greetings!

I have a performance question about mailman-2.1.6.

We run a mail server that serves some fairly large, active mailing 
lists (2k-3k members, a few hundred posts a day).  But we also have 
several smaller lists (20-30 members, with only a few posts a day).

We have been running mailman 2.1.5 successfully for quite a long time 
and have been very happy with it (thanks for all the work!).  However, 
we recently upgraded to 2.1.6:

- we used the same ./configure options for 2.1.6 as we did with 2.1.5
- our mm_cfg.py was unchanged between 2.1.5 and 2.1.6.

My back-end MTA is sendmail (not controlled by me -- nothing I can do 
about it).  It has not changed in configuration since we upgraded to 
2.1.6.

Previously, mail to my smaller lists would be delivered just about 
"immediately" (usually within a few minutes).  Mail to lists would flow 
through mailman quickly and be given to sendmail (which usually 
delivered it fairly quickly, particularly to local recipients).

However, once we upgraded to 2.1.6, performance on our smaller lists 
decreased dramatically during peak traffic times -- it can take 30-90 
minutes before mailman hands off incoming list mail to sendmail.  
sendmail still delivers the mail pretty quickly (particularly local 
recipients).  Specifically, outgoing mail files seem to sit in 
qfiles/out for a long, long time.  This only happens during peak 
traffic times -- during off-peak times, mail flows through mailman to 
sendmail quickly, and end-to-end deliveries are quite snappy (within a 
few minutes at most).

Does mailman somehow throttle/queue its output to the MTA?  Are there 
any configuration parameters that I can tweak to change this behavior?  
It's quite a bummer, especially since 2.1.5 did not seem to exhibit 
this behavior.  We see the parameter SMTP_MAX_RCPTS; it's listed in our 
Defaults.py with a default value of 500 (which seems to be fine).  I'm 
thinking that this is not our problem (it was 500 in 2.1.5 as well) -- 
indeed, the problem seems to be the rate at which mailman gives mail to 
sendmail, not how many mails it gives in each MTA transaction.

Any suggestions/comments would be appreciated.

Many thanks.

-- 
{+} Jeff Squyres
{+} jsquyres at osl.iu.edu
{+} Post Doctoral Research Associate, Open Systems Lab, Indiana 
University
{+} http://www.osl.iu.edu/