[Mailman-Users] Migration from Majordomo
Barry A. Warsaw
bwarsaw at cnri.reston.va.us
Fri Nov 26 05:05:39 CET 1999
>>>>> "Hrvoje" == Hrvoje Niksic <hniksic at iskon.hr> writes:
Hrvoje> It appears as if I should wait for a few months before
Hrvoje> installing it to a production machine. (But please Barry
Hrvoje> correct me if you think I'm wrong). Particularly, my
Hrvoje> setup would depend on stable config and list files, and
Hrvoje> smooth upgrades.
Let me say this: I'm definitely using Mailman in a production
environment -- every list on python.org is running off of it. But if
I were to be brutally honest, I'd say the 1.1 codebase still has
performance problems, which I now think are at least partly caused by
the number of Python processes that get created for each message
delivery. I've been doing a lot of work on all this stuff, and slowly
migrating our lists to the new codebase. I'm cautiously
Hrvoje> Config files not being plain text could be a problem.
I actually have a neat idea of how to handle this. If I get a chance
to code it up this weekend, I'll post a message about it (but probably
just to mailman-developers). If it works out, it'll let us do a bunch
of really cool stuff in this regard, including giving you plain text
config files you can edit.
Hrvoje> Speed also sounds like a potential problem. Python is a
Hrvoje> great language, but the current implementation is horribly
Hrvoje> slow, even compared to other interpreters. I have fears
Hrvoje> that other solutions, like Listar, will prove more
Hrvoje> feasible as the number of lists increases to 50 or 100+.
I haven't had time to look at Listar in much detail, but I'm not
concerned about Python's performance in all this, except in one
regard: the interpreter's start up time. But the new codebase creates
only one process per message so this should improve considerably
(getting the direct SMTP delivery module ported may change this, but I
will definitely attempt to keep this under control).
I agree with JC here, Python's performance /in general/ is no better
or worse than any other scripting language, and Python's other
advantages IMO far outweigh any of its problems. I still believe that
with sensible algorithms, your MTA is going to be a much greater
factor when considering performance (I'm alrady seeing awesome
performance from Postfix compared to Sendmail).
More information about the Mailman-Users