Here's how we set up our site-wide SpamAssassin bayes database, including the ability for users to train it.
Create a new user spamd with home /var/local/lib/spamd.
Add "spamd" to /etc/cron.allow.
Perform the following as spamd:
cd ~spamd
maildirmake -S Maildir, to create the shared SpamAssassin mailbox.
maildirmake -f SiteSpam -s write Maildir, to create a writable folder for misclassified spam (or if extracting from a tarball, make sure it has the sticky bit set by doing the following).
find ~spamd/Maildir/.SiteSpam -type d -exec chmod o+t {} \;
maildirmake -f SiteHam -s write Maildir, to create a writable folder for misclassified ham (or if extracting from a tarball, make sure it has the sticky bit set by doing the following).
find ~spamd/Maildir/.SiteHam -type d -exec chmod o+t {} \;
Add the following to ~spamd/.crontab to learn from and delete messages in those shared folders every five minutes (changing MACHINENAME to be the name of the local machine):
MAILTO=logs@MACHINENAME.hcoop.net # Learn from submitted spam 0,10,20,30,40,50 * * * * ~spamd/scripts/learn-spam --spam # Learn from submitted ham 0,10,20,30,40,50 * * * * ~spamd/scripts/learn-spam --ham # Remove any tmp cruft 3 3 * * * find ~spamd/Maildir/.SiteHam/tmp -type f -delete ; find ~spamd/Maildir/.SiteSpam/tmp -type f -delete
Be sure there's a newline after the last line, or it won't be processed.
Copy the learn-spam script from the spam directory of the hcoop "misc" repository into the directory ~spamd/scripts.
Modify /etc/spamassassin/local.cf with the directive:
# Location of bayes data bayes_path /var/local/lib/spamd/bayes/.spamassassin/bayes # Fix bayes permissions bayes_file_mode 0770 # Directives from old setup # [any custom stuff from the old /etc/spamassassin/local.cf that you want to keep]
Modify /etc/default/spamassassin by setting OPTIONS and ENABLEDas follows. The -x prevents spamd from trying to look for per-user configuration, which would be silly because it always runs as the same user here. Without this flag, the cron job triggered every 5 minutes would log an error message, which would lead to an e-mail being sent to the spamd user.
# Change to one to enable spamd ENABLED=1 OPTIONS="--create-prefs --max-children 5 --helper-home-dir=/var/local/lib/spamd -u spamd -x -s /var/log/spamd.log" PIDFILE="/var/local/lib/spamd/pid"
Make a file called /etc/logrotate.d/spamd with the following contents.
/var/log/spamd.log { weekly missingok create 0640 root adm rotate 4 compress delaycompress sharedscripts postrotate [ -f '/var/local/lib/spamd/pid' ] && (kill -HUP `cat /var/local/lib/spamd/pid`) || exit 0 endscript }
Start the daemon by doing /etc/init.d/spamassassin start. Check /var/log/spamd.log to be sure that it started OK.
Install the .crontab entries that you wrote earlier by doing crontab -u spamd ~spamd/.crontab as root. Do this every time that you make changes to ~spamd/.crontab.
Edit /etc/courier/shared/index as follows, being sure to separate each column with a single TAB character. The second column is UID, and third column is GID -- consult /etc/passwd and /etc/group to make these match the spamd user and group.
spamd 116 119 /var/local/lib/spamd
Restart courier's IMAP process: runsv restart courier-imap
Test by checking to see if you can access shared.SpamAssassin.SiteHam and shared.SpamAssassin.SiteSpam from IMAP. If not, do maildirmake --add SpamAssassin=~spamd/Maildir ~/Maildir as your normal user from the machine that does courier (and presumably spamassassin as well). You might need to replace ~ with ~USERNAME if you are using sudo to do this, where USERNAME is your normal username.
Now copy some spammy mail into the SiteSpam directory, wait 5 minutes, and check to see if the mail got learned and deleted.
If so, edit ~spamd/.crontab to pipe the output of sa-learn to /dev/null, and run crontab as specified earlier to propogate this change.