[Mailman-Users] A Known bug with confirmation string & excessive bounces

[Barry Warsaw]

On Mon, 2003-02-17 at 14:03, Tspeight at rockygorge.com wrote:
> I have sent this in before but I guess the powers that be are NOT 
> interested.  There is definitely something wrong with the bounce process 
> with regards to the Confirmation string.

Apologies for the delay in responding.

> The first time I reported this the problem occurred on the Mailman-users 
> list.  The other times it is occurring on a list that I host.  This is not 
> a random problem.
> 
> On 14 January 2003, I received the following message:
> "Your membership in the mailing list Mailman-Users has been disabled due to 
> excessive bounces The last bounce received from you was dated 14-Jan-2003."
> 
> On 13 February 2003, I received the following message: "Your membership in 
> the mailing list Mailman-Users has been disabled due to excessive bounces 
> The last bounce received from you was dated 14-Jan-2003."
> 
> The confirmation sting that was sent on both times follows.
> 37037b7a051f71a816e27e8e0b9ba82aaf107f7b
> 
> The confirmation string has been invalid each time!  Further on one of my 
> lists a member received a confirmation string but his account was okay.  Is 
> anyone else experiencing this problem?

Here are my suspicions.  I'm Cc'ing Thomas Wouters because he was
recently fixing some bugs in what I believe is the offending code.

First a question: for those of you seeing this problem, are your lists
on fairly active Mailman systems?  For purposes of this discussion, I'd
classify python.org as active enough.

The confirmation strings are kept in a database file shared by all lists
on your site, called data/pending.pck.  Normally, access to this file is
serialized via a lock file, but in we have our suspicions that there is
a race condition or other bug in the LockFile.py implementation that may
allow concurrent writes to this file.  We haven't been able to debug the
problem, nor is it easily reproducible, but if it's true, then my
suspicion is that you're getting bitten by the same bug.  I don't know
for sure, but I can say that in simple, one-by-one testing on my laptop,
I can't provoke an invalid confirmation string.

Now Thomas implemented some fixes to pending.pck which should avoid the
race condition, if one exists.  This is in cvs and will be part of
2.1.2.  python.org should be running this codebase so if you see the
problem on mailman-users or mailman-developers from here on out, I want
to know about it.

The real fix is for each list to get their own pending database, but I
don't think that'll get implemented in the 2.1-maint branch unless
Thomas's fix isn't enough.

-Barry