Hi Udo, While I am not 100% sure, I think that SpamAssassin's user based bayes system was not originally designed for virtual domains. That is, it was designed assuming that each machine would host only one domain and each mail user on that machine would have their own PAM account, directory structure and mail file. Therefore, it probably does not make alot of sense to try to do user based bayes reporting with most modern mail systems since most modern mail systems, such as Exim4U, use virtual domains. Any difference between a user based bayes system versus a common bayes system would depend on whether or not you were reporting spam to SpamAssassin with sa-learn. However, due to SpamAssassin's autolearn feature, I believe that there are benefits from using the bayes system with a common bayes database even if you are not reporting spam with sa-learn. Exim4U's recommended SpamAssassin configuration utilizes many other SpamAssassin checks such as the Open Protect Rules Channel and the checksum tests with Razor, DCC and Pyzor all of which help SpamAssassin accurately determine spam from ham. Therefore, the bayes autolearn system subsequently benefits from the overall SpamAssassin results in associating which tokens, character sequences and words are contained in spam mail versus ham mail as dictated by the many other SpamAssassin checks. I generally setup my installations to use a common bayes implementation with autolearn enabled. However, I am not a SpamAssassin expert so I would recommend that you pursue these questions further on the SpamAssassin mailing lists here: http://wiki.apache.org/spamassassin/MailingLists I hope that this helps. Gordon On 09/16/2010 01:58 PM, Udo Hortian wrote:
Hello Gordon,
thank you very much for your reply. The solution you proposed resolved the problem. On Debian the options for spamd are in /etc/default/spamassassin
But I am wondering if this is really a good solution. Does one want a common bayes database? Would not it be better to do this user-based or do deactivate it?
In http://exim4u.org/svn/exim4u_src/trunk/SPAMASSASSIN you write under point 6) about user specific spamassassin settings. But it seems that this does not really apply to exim4u, is not it? I think usually one would run exim4u with a single (or a few) user account, is not it? I mean not every virtual mail account will have an own system account.
Kind regards, Udo
_______________________________________________ users mailing list users(a)exim4u.org https://exim4u.org/mailman/listinfo/users