Bayes will not run auto/manual expiry

Matt Kettler mkettler at
Thu Jun 21 19:11:15 IST 2007

Douglas Ward wrote:
> I found an abort message in spamassassin (bayes related).  Does this
> look bad? 


> Has my bayes db grown too large to expire?

There is no such thing as too large to expire. My own bayes database won't even
consider trying expiry until it is larger than yours is currently. (I've
intentionally over-ridden the size, mostly for my own purposes.)

>  Should I clear it and start over?  

Probably not unless this becomes a persistent problem.

Any thoughts would be most helpful.  Thanks!

Looking at the results, the data in your bayes database is all "too close
together" in age.

SA does expiry based on how long it's been since a token was seen in a message.

Right now, if SA were to pick a cutoff such that it discarded the oldest 12
hours of tokens, your bayes DB would plummet to 73,442 tokens in size,
discarding 242,955 tokens in one blow.

>From the looks of it, you did a large "mass training" session about 71 days ago,
probably when you first set SA up. There's been some learning since, but not
much, or at least, not enough to act as a sufficient bayes DB of its own. That
initial burst still represents 76% of what's in your bayes DB.

I'd give it some fresh training if you have a chance, and try expiring again in
a few more days.

More information about the MailScanner mailing list