Im looking for ideas to make my spamfilter better. I catch a lot of spam, but I am still getting some spam in my inboxes and I also have some false positive spam (mostly thats because of razor btw). Its an old domain, and some of the email addresses are "exposed", they where on websites as plain text for long time Im currently using spamassassin with razor2 and the with following plugins: Mail::SpamAssassin:lugin:: SPF Mail::SpamAssassin:lugin:: TextCat I also use in my local.cf file some manual rules and the blacklist_from and whitelist_from configs. I am also running sa-learn as a daily cron job, to make spamassassin learn the spam and the ham mails. Sa-learn has been running for almost six month now, but I can not see any good or bad effects of the sa-learn. Can I do something to make it more effective maybe? I also tried the TxRep as an AutoWhitelisting plugin and I think it would be really good but unfortunately this plugin is not so maintained, and generated a lot of errors on my server, also almost doubled my server load. (I used it with mysql database as I was told it should be faster). I would prefer a good working solution for AutoWhitelisting, so spamassassin could learn the outgoing email addresses (and servers) and therefore avoid false positive spams, but I could not find a working and effective solution. I know there are no 100% solutions but there must be always a way to make things better. Any suggestions?
Do you run sa-learn under the user that runs amavisd on your server? If not, then it might be that it puts the learnt tokens into a wrong database (e.g. a database for the root user if you run it as root).
Thx for the tip, there really was some mixup, I am running amavisd under amavis user but the /var/lib/spamassassin folder was owned by debian-spamd so I had to do a chown amavis:amavis /var/lib/spamassassin -R because I had Code: bayes_path /var/lib/spamassassin/bayes/bayes bayes_file_mode 664 in my local.cf Also I was running sa-learn with -u vmail but now this will be -u amavis. I assume with this, sa-learn will have some effect.
Is there a way I can make sure sa-learn is working? I have a bayes_toks file thats 40M and my bayes_seen file is 10M big, but I never saw any entry in my mail headers that somehow relates to the fact that that score was given by sa-learn. Any suggestions for auto-whitelisting?
