[Mailman-Developers] Any other NNTP gatewayers out there? SMTP reject bounce processing is broken?

[Barry Warsaw]

Does anybody else gateway messages from NNTP, and specifically Usenet
using Mailman 2.1.x?

We've been seeing a very weird behavior where python-list (which is
gated to and from comp.lang.python) suddenly mass bounce disables every
regular delivery member.  The triggering bounce messages always come
from NNTP, never from email.  The problem doesn't appear to affect
digest members.  There are other clues <wink>.

Here's what I believe happens: Mailman pulls a message pulled from NNTP,
and tries to post it to the list.  The message however triggers and Exim
acl check such as the one on message header syntax or one of our
homegrown virus checks.  Exim then rejects the message at SMTP hand-off
time.  Mailman see that rejection and scores a bounce message for every
SMTP recipient.  Now, I VERP that list so everybody gets a separate SMTP
session, but I see about the number of rejects as the number of
non-already-disabled regular delivery members.  Exim never rejects a
digest containing one of the trigger messages.  Any message posted via
mail that would hit one of our Exim acls would never even make it to the
list, but NNTP pulled messages bypass that front-door check.

I think part of the problem is our Exim config, and I'm working with the
other python.org postmasters to address that.

The other problem is in Mailman's handling of rejections at SMTP time. 
Currently, it always scores a bounce for any recipients of such a
rejected message, and this scoring bypasses the bounce_processing flag. 
I think this behavior is at least partially broken.

One fix would be to never score a bounce if the message was rejected at
SMTP time.  We could move the message to qfiles/bad and log the
rejection but otherwise ignore the scoring.  This might be bad because
then Mailman wouldn't score bounces for non-existent local users, and
because on a VERP'd list, we might get tons of qfiles/bad messages for
every gated message.  This latter might be manageable.

We could just ignore rejections only for messages pulled from NNTP,
which should reduce the missing of legitimate scoring for bogus local
users, since they'll still get scored for regular email posts. 
Anybody have any thoughts or comments?  I'm probably going to implement
the latter to fix the immediate problem.

-Barry