On Thu, Aug 05, 2004 at 10:36:56AM -0400, Jeff Koch wrote: > > We tried the new Bayes SQL DB system (with new empty tables) and the load > was high. Then we tried using the site-wide Bayes DB that had been built by > 2.63 and the load was high. We also tried it with Bayes off completely with > no change so we turned that back on. Scans were taking as much as 30 to 40 > seconds. Server load averaged around 6.00. Now that we're back on 2.64 > scans are taking the more normal 1 to 2 seconds and the server load is > under 1.00.
If you run spamd with -L how do things look? Note: you can't downgrade your bayes dbs, so once you upgrade them to version 3 you have to use them only with 3.0.0 and above. To go back to 2.64 you'll have to remove your bayes files. > One thing I noticed was that the maillog showed more detail with 3.0 - it > showed all the tests being done. Maybe the SURBL tests take too long or > require too much memory. Also, is it possible that SPF is automatically > enabled? > If the proper modules are installed then yes SPF will run. The SURBL tests will also run if all the right pieces are in place. I suggest you start turning things off (ie -L, use_bayes 0, use_auto_whitelist 0, etc) til you figure out what exactly is making things slow. My bet was on bayes upgrading the database files, that would explain the high CPU and memory. If you wipe out or turn off your bayes and the CPU/memory requirements go down but you're still slow then I would put money on a slow DNS server or network in general. Lots of people have been running the 3.0.0 code in one form or another for a very long time so it's hard for me to believe it's something code related Not that it couldn't happen, but it's important at this point to do as much debugging as possible to find the primary cause. Michael
