On 13 Aug 2015, at 19:59, Joe Acquisto-j4 wrote:
Last few days, noticed getting two of some messages. Been busy at my
day job and brushed it off. But now it appears to be happening with
some (ir)regularity.
I can see from /var/log/mail that the repeat messages do have
identical message-id. The only difference that caught my bleary eye
was that for one
bayes=0.000000,autolearn=ham
for the other
bayes=0.000000,autolearn=unavailable
Is the cause so obvious I should know already? Hints? Ready fix?
spamassassin --version gives
SpamAssassin version 3.3.2
running on Perl version 5.16.0
Sure, fine, but that's probably irrelevant. You don't mention how you're
using SpamAssassin but it is almost never capable of causing message
echoing, because it is almost always used in a consultative manner by
something else that is responsible for storing and transporting mail: a
MTA, a SMTP filtering proxy, etc. SA never has full custody of a
message. You've mentioned no software that is responsible for transport
or delivery of messages, so giving you anything but vague guesses is
impossible
One common cause of receiving 2 copies of the same message is a network
problem that occasionally drops outbound small packets. The last step in
passing a message via SMTP is the receiving mail server sending an
acknowledgement of having received the message OR an error reply telling
the sender that it could not accept the message, either for a reason
that should be deemed permanent or one that could be resolved in the
near future. If that response is lost, the sender will eventually retry,
while the receiver happily does its job of delivery. When the retry
happens, the message is duplicated. This can also happen if your MTA is
waiting for SpamAssassin to give its opinion before responding, and the
sender times out in some way that your MTA can't detect. That can be
caused by overaggressive firewalls, most notoriously Cisco gear with
their spectacularly bad SMTP 'fixup' misfeature.
I recently had a spate of such duplicates when my ISP tried to roll out
a change on their gateway devices that tried to enforce a 1:1 mapping of
IPs to MAC addresses, even for business customers like me with multiple
static IPs allocated. About a dozen times a day my MTA would get shunned
by my next hop just long enough for the sender to be told it was
unreachable while the MTA was waiting for something more after that last
message that never came. It would eventually time out the SMTP session
thinking it had received and acknowledged a piece of mail, and the
sender woulds retry, thinking my MTA had died without receiving the
mail.
The reason the two messages got two different autolearn results is
possibly related (e.g. if there's a problem with your Bayes DB that is
making SA take a long time to score messages or give up) but without
knowing more details (Sendmail? Postfix? Amavis? MIMEDefang? BayesDB on
a MySQL server halfway around the world? An MTA living behind a PIX that
hasn't been touched in a decade?) anything along those lines is open to
unrestricted imagination.
