On 12/16/15 15:52, Steve Shockley wrote:
I recently ran into an issue with my OpenBSD mail server where it would die every day around 5 AM. With 5.7-stable it would just become unresponsive, with 5.8-stable it would print "scsi_xfer pool exhausted" repeatedly on the console. It turned out to be SpamAsssassin sa-learn running on a folder with 26k+ messages, and I simply ran out of memory and swap.
Have those messages accumulated over some time, or do you receive that number of messages within a very short time frame?
I'm asking because unless I remember this incorrectly, there's no need to have sa-learn run more than once on any given message (the tokenized form is stored in the /var/db/spamassassin/bayes hierarchy). If you have a cron job or similar running sa-learn, it's likely useful have the scrimp move already scanned messages off to somewhere out of the way or remove them entirely.
That said, while I haven't had spamassassin chrash machines yet, I've occasionally seen spamassassin's perl processes consume 100% cpu for long periods on slow machines, leading to spamassassin timeouts and delayed delivery. Emptying out the sa-learn input directories usually helps, and of course some version of throwing more or faster hardware at the problem might also help.
-- Peter N. M. Hansteen, member of the first RFC 1149 implementation team http://bsdly.blogspot.com/ http://www.bsdly.net/ http://www.nuug.no/ "Remember to set the evil bit on all malicious network traffic" delilah spamd[29949]: 85.152.224.147: disconnected after 42673 seconds.
