SA Bugzilla – Bug 2748
Message is passed through without being processed
Last modified: 2004-02-18 10:32:44 UTC
A message came through with no spamd header. I did some research and found the following in my logs: Nov 11 07:58:50 argos spamd[33462]: Cannot open bayes databases /home/gostl/.spamassassin/bayes_* R/W: lock failed: Do not specify an existing file. I believe this was caused when two messages came in simultaneously. We see 10- 20 of these each day for app 8,000 emails scanned. I suspect you need to somehow serialize spamd on a per user basis. If it were in C, I'd consider a semaphore of some kind. My perl isn't good enough to make a creative suggestion there.
Hi -- we do serialize access and manage the concurrency issues; that message is informative only, noting that the bayes db could not be written to (it means that SA tried to auto-learn a message, but another process had the file locked). parallel scanning works fine, as long as the load isn't too high; if the load is high, there'll be more requests than -m permits, and the other requests will be queued. If they stay in the queue too long, they'll be skipped and passed on unscanned. So that could be what happened there, I'd reckon. I'd suggest looking through syslog for lines related to that Message-ID, and see if there's any error messages from spamc.
Subject: Re: Message is passed through without being processed I've lost track of the original message, so I can't do much more checking, but other than the locking error, there were no spamc/spamd error messages. No max-children messages anywhere around there. On Tue, 11 Nov 2003 bugzilla-daemon@bugzilla.spamassassin.org wrote: > http://bugzilla.spamassassin.org/show_bug.cgi?id=2748 > > > > > > ------- Additional Comments From jm@jmason.org 2003-11-11 11:08 ------- > Hi -- > > we do serialize access and manage the concurrency issues; that message is > informative only, noting that the bayes db could not be written to (it means > that SA tried to auto-learn a message, but another process had the file locked). > parallel scanning works fine, as long as the load isn't too high; if the load is > high, there'll be more requests than -m permits, and the other requests will be > queued. If they stay in the queue too long, they'll be skipped and passed on > unscanned. So that could be what happened there, I'd reckon. > > I'd suggest looking through syslog for lines related to that Message-ID, and see > if there's any error messages from spamc. > > > > ------- You are receiving this mail because: ------- > You reported the bug, or are watching the reporter. >
I haven't seen that specific bayes error ever, but I do occassionally get messages w/out headers. This is always due to an overloaded server, and procmail eventually times out the spamc call. The Bayes lock failure wouldn't cause a spamd abort, so I'm assuming the spamc call was timed out. If this issue is reproducable, and it's not a timeout issue, we can reopen. :)