Jack Gostl wrote: > I've been looking at the spams that slip through, and I notice that > they have no Bayes score. Not a low score, but no score. I suspect > that is tied to this message in my log: > > Dec 2 02:00:44 web01 spamd[21664]: bayes: cannot open bayes > databases /home/gostl/.spamassassin/bayes_* R/W: lock failed: A > system call received an interrupt. > > I'm guessing that this is the result of two copies of spamd hitting > the Bayes files at once. Since we have several people sharing the same > Bayes files, this is a distinct possibility.
R/W lock failures should not have amy impact on scoring. It will just cause autolearning to be skipped. Now R/O lock failures WILL cause bayes to not be used in scoring, but you're not getting those.
