> I've been on lists that went through duplication periods, where
> every post came through twice, and now it's happening on one of my
> lists (all lists in question are majordomo run). What causes
> duplication and how does one fix it? I've checked the headers of
> some duplicated messages to see if they're coming to slightly
> diff. addresses, and that isn't it.
I've seen several reasons for duplicate messages.
1. It is normal for an occasional message to be duplicated. For
example, machine A completely transfers a message to machine B but
never gets a response from B that B got the message. So machine A
sends it again. It's rare that such a response gets lost (or that
either machine A or B shuts down at just the right time), but it does
happen.
2. Extraordinary circumstances can cause the above to happen more
often that it should. For instance, some mailers won't return the
response that "they got the message" until they've tried to send it to
every one of the recipients of that message (or at least verify their
addresses). For a long list, this can take so long that the sending
mailer (machine A in the above example) assumes that machine B has
crashed, so machine A re-sends the message. Because it's affected by
network delays, the symptom can come and go with no apparent cause.
3. Some thoroughly broken mail gateways to LAN environments will,
instead of bouncing a message to the sender, re-send a message to each
of its recipients. (One of these just sent 600+ duplicate messages to
one of my lists.)
4. The same thing can occur with gateways between mailing lists and
Usenet newsgroups, if they aren't set up carefully. The newsgroup
sends a message to the mailing list, which promptly sends it back to
the newsgroup. If the message-id changes (some lists do this), Usenet
will think it's a new article and send it back to the mailing list.
You can usually tell where the duplication is coming from by comparing
the Received headers of a few copies of such a message, and see where
they diverge. (The dates on the Received headers of two copies of a
duplicated message will be the same before the point of duplication,
and different afterward.)
Some list managers try to be smart and detect duplicated messages by
recognizing any text and/or message-id that appears more than once.
Sometimes this works very well; at other times it ends up rejecting
perfectly valid messages that happen to contain parts of old messages.
(In particular, Listproc 6.0 is notorious for doing this badly.)
Keith
References:
|
|
