Am Fre, 2004-06-25 um 18.00 schrieb Michael Parker:
> On Fri, Jun 25, 2004 at 05:43:54PM +0200, Harald Arnold wrote:
> > I have all my mails beginning from year 2000.
> > I tried to feed the database (ham <-> spam)
> > via sa-learning --[ham|spam].
> > After learning about 2.000 mails I tried
> > to see the "magic" numbers (nham and nspam),
> > but the database was KILLED and destroyed:
> 
> Was anything else hitting the database at the same time (ie scanning)?
> 
> It's possible that your sa-learn took so long that the scanning code
> thought that the lock was stale and broke the lock and started writing
> to the database.  This could/would cause corruption in the database.
> This has been fixed in 3.0.0 by refreshing the lock every so often.

This IS possible. Today I wrote two articels that I always got the
message 

"Argument "\060372822���@" isn't numeric in numeric lt (<) at
/usr/lib/perl5/site_perl/5.8.1/Mail/SpamAssassin/BayesStore.pm line
1267.

an afterwards the count of learned spams decreased (sometimes to 0 or
3 although I taught hunders of spams).

Another problem was when using "sa-learn --rebuild --force-expire":

debug: bayes: found bayes db version 2
debug: bayes: expiry check keep size, 75% of max: 112500
debug: bayes: token count: 225055, final goal reduction size: 112555
debug: bayes: First pass?  Current: 1088164514, Last: 1088164472, atime:
0, count: 0, newdelta: 0, ratio: 0
debug: bayes: Can't use estimation method for expiry, something fishy,
calculating optimal atime delta (first pass)

Thanks Harald

Reply via email to