MailScanner + Spamassassin 3.13 autolearn not working !
Joni Bäcklund
joni at connect.fi
Sun Jul 2 19:04:27 IST 2006
Hello Friends !
I have been running SA for years with mailscanner and some virus
scanning software on our
mail linux fedora server. This server now runs FC5 and I found out
some two weeks ago
that bayes database was somehow corrupted so I did go for
fresh mysql bayes database and learned some thousands of saved ham and
spam messages.
Everything seems to be on order now and I see bayes hits on maillog
as usual.
The problems seems to be that SA never does autolearn even the spam
value is very high (
like 15 or so ). So the bayes data stays permanent if I do not
train some messages manually by sa-learn.
I have tested also the system so that I used the normal bayes DB on
/var/spool/spamassassin instead of mysql with same bayes data ! Same
behavior !
I never see on the maillog the autolearn=spam ham or disabled. No
autolearn markings at all ! The Autolearn
module is loaded and everything else seems to be running ok. The same
happens with
autowhitelist. No entries are made to auto-whitelist ! That seems
strange. The autowhitelist file itself is at /var/spool/spamassassin
but its allways same size and does not grow !
Im running spamassassin 3.1.3
perl-5.8.8
mailscanner-4.55.7-1
and here are some lines from spam.assassin.prefs.conf that mailscanner
uses with SA.
#
# JKF 28/04/2003
# The following settings has been pretty much superceded by the "Advanced
# SpamAssassin Settings" in MailScanner.conf.
#
# JKF 26/03/2003
# If your root filesystem is filling up because SpamAssassin is putting
# large databases in /.spamassassin or /root/.spamassassin, you can move
# them using the following lines to point to their new locations.
# The last part of the path is not a directory name, but actually the
# start of the filenames. So with the settings below, the Bayes files will
# be created as /var/spool/spamassassin/bayes_msgcount, etc.
#
auto_whitelist_path /var/spool/spamassassin/auto-whitelist
auto_whitelist_file_mode 0600
use_auto_whitelist 1
#bayes_path /var/spool/spamassassin/bayes
#bayes_file_mode 0600
bayes_store_module Mail::SpamAssassin::BayesStore::SQL
bayes_sql_dsn DBI:mysql:sa_bayes:localhost
bayes_sql_username XXXX
bayes_sql_password XXXXX
bayes_sql_override_username XXXX
# Most people don't use NFS-shared Bayes databases
# so this is added for SpamAssassin 3
lock_method flock
required_hits 4.5
bayes_auto_learn_threshold_spam 8.0
bayes_auto_learn 1
score BAYES_99 0 0 4.070 3.601
# Most people don't use NFS-shared Bayes databases
# so this is added for SpamAssassin 3
lock_method flock
The spamassassin -D --lint works without any problems and all required
modules are loaded !
Are there some undocumented changes in autolearn behavior with mailscanner ?
here is sa-learn --dump magic output
0.000 0 3 0 non-token data: bayes db version
0.000 0 5486 0 non-token data: nspam
0.000 0 7142 0 non-token data: nham
0.000 0 187972 0 non-token data: ntokens
0.000 0 1036499585 0 non-token data: oldest atime
0.000 0 1151841259 0 non-token data: newest atime
0.000 0 0 0 non-token data: last journal
sync atime
0.000 0 1151833489 0 non-token data: last expiry atime
0.000 0 22118400 0 non-token data: last expire
atime delta
0.000 0 16694 0 non-token data: last expire
reduction count
So I just do not understand what is going wrong there as the SA with
mailscanner system
has been workin for years and autolearn etc. has been working
nicely too !
Joni
--
Joni Bäcklund, Tel +358400665775, FAX +35898042007
Email: joni at connect.fi, oh2njr at sral.fi
Amateur packet radio AX25: oh2njr at oh2rbj.fin.eu
Some kind of Homepage: http://www.connect.fi/joni
" The Choice of a GNU generation: SuSE Linux 10.1 "
-------------------------------------------------------------------------------
This mail sent through Connect Services WebMail : https://webmail.connect.fi
More information about the MailScanner
mailing list