Hi

I'm having some troubles on a high-traffic MTA. The mailserver runs
sendmail and maildrop, configured to scan messages during the local
delivery, using spamc, against a different server running spamd.

I wish to have all local mail scanned, but if there is a problem with
SA, or SA is taking too long, have the message delivered locally
without being scanned. I do not want maildrop to exit with a
EX_TEMPFAIL when SA fails.

So spamc is invoked from maildrop with:
 xfilter "/usr/bin/spamc -d 1.1.1.1 -f -t15 -s150000"
(-f to continue if something fails, and -t15 to wait 15sec max.)

On the SA server, I have spamd running with:
 spamd -d -u spamd -i -s local4 -m15 -x -r /var/run/spamassassin/spamd.pid

It all works great, unless we have a spike on the amount of mails/sec
handled, which usually happens once or twice a day. When the load goes
up, I get a large "queue" of spamc processes (around 140 this
afternoon), all waiting to be scanned by spamd. During that time, some
of them takes more than 5 minutes, and then maildrop cancels and exits
with a EX_TEMPFAIL error. This causes sendmail to queue the message
for around 30 minutes, and then try local delivery again:

Mar 15 12:45:58 vmail-av1 sendmail[7232]: k2FFhvrA007204: timeout
waiting for input from local during Draining Input
Mar 15 12:48:58 vmail-av1 sendmail[7232]: k2FFhvrA007204:
to=<[EMAIL PROTECTED]>, delay=00:05:01, xdelay=00:05:01, mailer=local,
pri=36466, relay=example.com, dsn=4.0.0, stat=Deferred: local mailer
(/usr/bin/maildrop) exited with EX_TEMPFAIL

(note the xdelay=00:05:01).

So it seems like spamc is not honouring the "-t 15" flag when invoked
from maildrop.

But when that happens, I tried to to a manual scan of a message from a
file, using -t1 for example, and then spamc exits outputting the
messages without being scanned, just like I would like it to behave
from maildrop.

I'm using SA 3.0.5 on a CentOS system. I saw the release notes for SA
3.1.1, which includes the resolution for bug ID 3710, related to the
-t flag, but I'm not having the problem described on that bug. From my
experience, SA seems to honour the -t flag when invoked from command
line.

Any clue? Should I invoke spamc differently? Should I upgrade to 3.1.1?

Regards.

Reply via email to