[Mailman-Users] Subscribers suddenly "disappear"

[Mark Sapiro]

Barry Finkel wrote:

>Max Lanfranconi <Max.Lanfranconi at Sun.COM> wrote:
>
>>Mailman 2.1.11
>>Python 2.4.4
>>OS Solaris 2.11
>>
>>
>>Hi,
>>
>>I have been able to reproduce this bug consistently by running the
>>replicate_bug script:
<snip>
>
>I ran the script (after some minor modifications) on
>
>     Ubuntu Dapper
>     Mailman 2.1.11  (self-built package)
>     Python 2.4.3 (#2, Oct  6 2006, 07:49:22)
>
>and I get similar results:
>
<snip>
>
>I then added
>
>     sleep 5
>
>after each "add_members" line, and the output looked fine.
>I changed the sleep interval from 5 down to 1 in successive
>runs, and each output looks fine; each list has the proper two
>subscribers.  Is there a timing issue here?


There appears to be some kind of race issue. There are two other
reports of strangeness like this at
<http://mail.python.org/pipermail/mailman-developers/2008-May/020190.html>
and
<http://mail.python.org/pipermail/mailman-users/2008-May/061720.html>.

In both of those cases it seemed there may have been a locking failure
of some kind, but I couldn't see how. See the above threads for more
detail.

However, locking (at least list locking at the Mailman level) doesn't
seem to be involved in this case because each command process will
have saved and unlocked the list and terminated before the next
command process begins.

I intend to test the script on two other platforms (Cygwin/Python
2.5.1, and Centos 5/Python 2.4.3) later today.

Given that this script is (or should be) running the command processes
serially with no contention between commands for the config.pck files,
It would seem like an OS issue, but the script failure occurs on two
different OS platforms already, so I'm at a loss for what the issue
may be.

-- 
Mark Sapiro <mark at msapiro.net>        The highway is for gamblers,
San Francisco Bay Area, California    better use your sense - B. Dylan