[Mailman-Users] Mailman causing strange slowdown in SMTP

[Craig Pettersen]

Thanks for the response.  The slow down has continued and nothing I've tried
has helped.  I answer the questions put to me in context below.  Any further
ideas would be greatly appreciated.


>>Hi,
> >>  I have a mailman installation running on our qmail server that is
> involved in a >>slowdown in SMTP for the designated SMTPHOST in mm_cfg.py.
> I first noticed >>that the messages for the list go out very slowly...about
> 15 seconds per >>message.  I saw this in the mailman smtp log, and then I
> saw it in the the qmail >>smtp log as well, so I figured it was a problem
> with qmail.  As I looked through the >>qmail logs, I saw that all messages
> that connected to smtp via 127.0.0.1 were >>slow, averaging more than 15
> seconds, while those that connected from >>somewhere else were all a
> fraction of a second.  I discovered that if I changed >>SMTPHOST to our
> public IP in mm_cfg.py, any messages (from the list or >>elsewhere) using it
> would slow down as well, and any messages connecting via >>127.0.0.1<http://127.0.0.1/>returned to normal speed.  If I shut down mailman using mailmanctl >>stop,
> speed was normal for both localhost or the public IP, whichever had been
> >>used as mailman's SMTPHOST at the time returning to mormal speed.  That
> >>verified that mailman is directly involved in the slowdown in smtp.  If I
> telnet to the >>smtp and manually send a message, the hangup point is after
> entering 'data' - ie:
>
> >>helo test.com
> >>250 ok
>
> >>mail from: test.com
> >>250 ok
> >>rcpt to: <a valid email>
> >>250 ok
> >>data
> <-------after entering 'data' there is a 15 second + wait if mailman is running
>
> >>354 ok
> >>etc...
>
>
>
>
>
> >>This mailman installation serves only one list, an announcement only
> > list,
> > >>of about 35,000 members, with over 10,000 disabled due to bounces that
> > send
> > >>once every 2 weeks or so. It invariably sends
> > >>several hundred to a thousand messages individually at the
> > +/-15sec./message
> > >>rate, then a big bulk mailing to most subscribers that's taking about
> > 3000
> > >>seconds for around 23,000, then a few score more of individual
> > messages
> > >>before it stops functioning.
> >
> >
> > >What exactly is in Mailman's smtp and smtp-failure logs (the
> > >'individual' messages are not the post).
>
>
>
> I figured out that the individual messages were messages left in
> qfiles/out.  When message are left like that they take around 15 seconds
> each to clear out. So this is reflected in the smtp log.  There are no
> unusual failures in smtp-failure.  It does appear from going through the
> logs that the problem happens when there are messages in qfiles/out.  When
> the list message goes out to the large list, qfiles/out begins to accumulate
> messages.  Those messages are taking the 15 seconds to go out.  Once
> qfiles/out clears up, the time appears to drop to normal, usually less than
> a second.
>
> >>Using ps aux, one can see that all the
> > >>qrunners are there and appear normal, but bin/mailmanctl must be used
> > to
> > >>stop and start things to get them going again.
> >
> >
> > >When things are not going, what is the contents of the queues
> > >(subdirectories of the qfiles/ directory)?
>
>
> So there are hundreds of messages backed up in qfiles/out, again each
> taking about 15 seconds , the vast majority not varying more than a second
> from that.
>
> >>  So my questions are...does anyone have any idea what mailman could be
> > >>doing to interfere with smtp on its SMTPHOST and how to fix it? and
> > does
> > >>anyone have any idea why mailman stops working shortly after sending
> > out the
> > >>big list?
> >
> >
> > >What are your various mm_cfg.py settings if any for the VERP settings
> > >and for SMTP_MAX_RCPTS and SMTP_MAX_SESSIONS_PER_CONNECTION.
> >
> > >You might try
> >
> > >SMTP_MAX_RCPTS = 10
> > >SMTP_MAX_SESSIONS_PER_CONNECTION = 1
> >
> > >To see if that makes a difference.
>
>
> I tried these settings and they don't make any difference.  I have no VERP
> settings in mm_cfg.py
>
> >If you restart both Mailman and qmail, does the first subsequent
> > >Mailman SMTP session exhibit the 15 second delay?
>
>
> If  I restart everything, invariably the first mesages is under 2 seconds
>
> >Have you seen
> > >< http://www.python.org/cgi-bin/faqw-mm.py?req=show&file=faq04.011.htp
> > >?
>
>
> I have looked at that.  I see that my concurrency for qmail is at 400.
> Mostly the points there seem to be ways to get better performance on a
> normally functioning list.  I would like to try some of them but my main
> problem is this unusual slowdown.  Everything was working fine a few eeks
> ago.  One strange thing about this is that the logs show the time
> consistently at about 50ms above 15 seconds.  Once in awhile I see one that
> is about 50ms about 14 seconds. Here for ex. are the last twelve times-
> 15.058, 15.056, 15.051, 15.060, 15.055, 15.055, 15.051, 15.051, 15.054,
> 15.054, 15.055, 15.051, 14.055.  This is very representive - almost like
> there's a fifteen second timer.  But every 12-15 or so there is one that's
> 14.0** .  Then, to make things interesting, every once in a while (after
> perhaps several hundred or more at 15.0**) there's one or more at a higher
> time - ranging up to about 120 seconds and then back to 15 sec.
>
> Thanks again...would appreciate any further ideas..
> -Craig Pettersen
>
> --
>
>
> Mark Sapiro <msapiro at value.net >       The highway is for gamblers,
> San Francisco Bay Area, California    better use your sense - B. Dylan
>
>
>