Just in case someone runs into something similar in the future... restoring the DB files from backup (and running sa-learn --sync once they were in place) seems to have done the trick. The main question that remains outstanding is whether or not the "best practice" is to do things this way (using a script to tar/zip the bayes files and then dump them to tape [as part of a regular backup]) or if using the sa-learn --backup > file is better, or does it not matter?
Thanks, Forrest ------ Original Message ------ Received: Wed, 29 Aug 2007 10:30:03 AM EDT From: "FH" <[EMAIL PROTECTED]> To: <users@spamassassin.apache.org> Subject: SA locked up and seemingly dumped the DB we've been building > When I got in Monday morning our mail server was almost completely locked up. > I wasn't able to login remotely, I was eventually able to login on the > terminal but that was only after pulling the network connection and giving it > a little time. Once I got on I found a bunch of out of memory/swap space > errors and top showed something like 90%+ CPU dedicated to User processes w/ > spamd being the largest contributor, so all that probably explains the lock > up. BTW the machine is an old Sun that is about to be replaced so I'm not too > worried about this past performance, and this has happened once or twice in > the past. When this has happened in the past what I've done is kill (I try to > do it gracefully) the spamd process and restart it. This time however after > restarting it it seems like we are almost back to the beginning wrt the bayes > DB. What I mean is messages that were getting marked properly on Sunday were > no longer getting marked properly on Monday (this is just going by subject > lines which I have a record of) and the performance/training metrics I keep > went from about 10-15% missed to 20-30%+ missed. A lot of the scores are > coming up as 2 (4 is my threshold for marking something as spam) which again > seems to be like it was when we first started IIRC, and some are even coming > up as negative scores now :( > > Has this happened to anyone else before, what's the best way to recover? I > should have a copy of the files in the bayes_path folder on a backup that was > done last week. Could it be as simple as putting those file back in place and > restarting spamd[1], might that get some of the info back? Is there any way > to verify what exactly happened? I searched the log files and didn't see > anything unusual reported by spamd (missing/corrupt DB errors or the such). > Are there any other commands I can run to double check that the current DB is > ok, maybe it just needs to be reindexed or something? > > Thanks, > Forrest > > > [1] On a related note, when I setup the new machine, can I install > spamassassin and then copy these files over the default files that are > created? Will that essentially move the bayes DB to the new machine? > > >
