[Mailman-Developers] Re: Extraordinary locking mechanisms...

[bwarsaw@python.org]

>>>>> "JH" == Jeremy Hylton <jeremy@cnri.reston.va.us> writes:

    JH> Chris pointed at that mailman runs as a distributed system, so
    JH> IPC won't work.  The idea, though, of using something other
    JH> than the filesystem for locking makes good sense.  The locking
    JH> could be based on a server that communication with other
    JH> mailman components over a socket.  It would look basically
    JH> like the Javaspaces leasing service.  (Leases are basically
    JH> locks that timeout.)  The only worry I would have is about
    JH> crashes of the lease service.  For a single machine, it could
    JH> run under init; for multiple machines, a leader election
    JH> protocol would be necessary to guard against machine crashes.

We also want to think about how all this will change if we move to a
monolithic daemon model, as opposed to the one-shot CGI model we're
currently using.  We may not need external locks at that point at all
(although we'll of course still have to guard against crashes of that
daemon).

-Barry