On Thu, Aug 05, 2004 at 08:26:37AM -0400, Jeff Koch wrote: > > I just installed this development release on one of our mailservers. It has > some great new features but it almost killed the machine. The load average > tripled, CPU usage went way up, time to scan a message went from an average > of 3 to 20 seconds, swap usage went to 500MB, etc. >
Not much to go on, but I do have a theory. In cases like this interactive debugging is often better, if you are still having problems log into IRC later and ask for help. Are you using a sitewide bayes db? or do you have individual user dbs? My theory is that SA is trying to upgrade your bayes db(s) all at once. You can force this manually (see UPGRADE) for each bayes db and then restart and things may be just fine. If you wanted to test the theory quickly with little to no work, just set use_bayes to 0 in your local.cf file and restart 3.0 processing. If you have load problems then that probably wasn't it. Here is an untested tip for doing this on a running system: 1) Install 3.0 1) Set use_bayes_rules to 0 in local.cf 2) Set bayes_auto_learn to 0 in local.cf 3) Start/restart spamd 4) Run sa-learn --sync on all of you bayes dbs 5) Unset the user_bayes_rules and bayes_auto_learn settings. 6) Start/restart spamd If it turns out that this solves the problem, can you let me know, I might want to beef up the UPGRADE documentation a little to help other folks. Michael
