Re: spamd and the nobody home

[Bob Proulx]

Bob Proulx wrote:
> I decided to fix this.  But don't want to use those files from 2012
> and 2013 so I move the directory to old-nobody and set up new files.
> I will experiment around to see if I can make this usable so that
> there can be Bayes learning for it.

I might have this working now.

    /var/lib/spamassassin/nobody:
    -rw-------  1 nobody       nogroup          6 Jul 18 21:06 bayes.mutex
    -rw-------  1 nobody       nogroup      12288 Jul 18 21:06 bayes_seen
    -rw-------  1 nobody       nogroup      24576 Jul 18 21:16 bayes_toks
    -rw-r--r--  1 root         root           571 Jul 18 20:42 user_prefs

For reasons of general performance in the user_prefs file I turned off
bayes_auto_expire.  That's a normal thing for me because inline
expiration takes a long time for normal message processing.  This
speeds that up significantly.

    # Because there is no NFS involved we can speed things up by specifying
    # to use flock explicitly.
    lock_method flock

    # Stop SA from running bayes expires (takes long time) during message
    # handling.  But now must have cronjob run sa-learn --force-expire!
    bayes_auto_expire 0

    # Default is 150000 tokens which on this busy system is one day.
    bayes_expiry_max_db_size 500000

That required setting up a separate cron job that is run every few
hours.  I also increased the size of the database as I have found that
helps with things too.  Here is the new /etc/cron.d/spamd-local file.

    30 0,6,12,18 * * *      root    start-stop-daemon --start --user nobody 
--pidfile /var/run/spamd.pid --name spamd --exec /usr/bin/sa-learn -- --dbpath 
/var/lib/spamassassin/nobody --force-expire

I used start-stop-daemon to switch user for consistency with other
things on this Debian based system.  Also it checks pid and name too.
If it were on a non-Debian system then I would use "su" for the same
purpose of switch user instead and not worried about the pid and name
checking.

After just a few minutes I see that messages are being processed.

    root@debbugs:~# start-stop-daemon --verbose --start --user nobody --pidfile 
/var/run/spamd.pid --name spamd --exec /usr/bin/sa-learn -- --dbpath 
/var/lib/spamassassin/nobody --dump magic
    Starting /usr/bin/sa-learn...
    0.000          0          3          0  non-token data: bayes db version
    0.000          0          0          0  non-token data: nspam
    0.000          0          4          0  non-token data: nham
    0.000          0        527          0  non-token data: ntokens
    0.000          0 1626656201          0  non-token data: oldest atime
    0.000          0 1626656774          0  non-token data: newest atime
    0.000          0          0          0  non-token data: last journal sync 
atime
    0.000          0 1626657419          0  non-token data: last expiry atime
    0.000          0          0          0  non-token data: last expire atime 
delta
    0.000          0          0          0  non-token data: last expire 
reduction count

Time to let it sit and simmer for a while and then check it again
later.  It needs to collect at least 200 messages of both nspam and
nham in order for SpamAssassin to use it.  I am sure that will happen
within a short time.  And then hopefully the Bayes will improve the
anti-spam accuracy somewhat.  Although frankly I am not sure where
this comes into play in the email flow of debbugs.  Maybe just exim
incoming and not elsewhere?  I have no idea.

Bob