On Wed, 29 Nov 2006 08:22:13 -0600 Bookworm <[EMAIL PROTECTED]> wrote: > It sounds like you have the spamd bayes database, and then you have > the database for whatever user you're actually running the test > from. I ran into this problem as well - it's a known issue, and I > wish the SA folks would come up with a way to run, as root, sa-learn > for a NON-ROOT bayes database. Vpopmail directories aren't readable > by spamd.
I'm not running vpopmail on this server. spamd is running as qscand, however I've got my /etc/spamassasin/local.cf set to use a site wide bayes database. I also have auto_whitelist configured, and I'm wondering if this is such a good idea: bayes_path /etc/mail/spamassassin/bayes bayes_file_mode 0770 auto_whitelist_path /etc/mail/spamassassin/auto-whitelist auto_whitelist_file_mode 0770 use_bayes 1 bayes_auto_learn 1 I have a script that runs every night that sa-learn's data from each users SpamTrain folder into this site wide database. Now I seriously hope that spamd isn't reading its bayes data from qscands home, as this data hasn't been touched for 2 years: ls -alh ~/qscand/.spamassassin drwxr-xr-x 2 qscand qscand 4.0K 2004-07-07 11:01 . drwxr-xr-x 4 qscand root 4.0K 2006-07-26 11:28 .. -rw-r--r-- 1 qscand qscand 20K 2004-01-06 16:43 auto-whitelist -rw------- 1 qscand qscand 556K 2004-07-07 11:01 auto-whitelist.dir -rw------- 1 qscand qscand 556K 2004-07-07 11:01 auto-whitelist.pag -rw------- 1 qscand qscand 47K 2004-07-07 11:01 bayes_journal -rw-r--r-- 1 qscand qscand 10M 2004-07-07 11:01 bayes_seen -rw------- 1 qscand qscand 4.2M 2004-07-07 11:01 bayes_toks -rw-r--r-- 1 qscand qscand 1.5K 2006-07-06 09:53 user_prefs Maybe I should delete that and symlink in the files to the sitewide bayes, just in case? When I ran spamassassin on the item of spam I referred to earlier, it was using my own user account. I don't even have any data in my .spamassassin folder, so I can only assume that it was using the site wide bayes for its checks then. Or could my bayes data be completely messed up, and spamassassin was doing a better job of identifying spam under my user account with no bayes data at all? Once again many thanks for the suggestions and help... Wilb
