[Mimedefang] problems with bayes (?) database access

[Kelson Vibber]

Dirk Mueller <dmuell at gmx.net> wrote:
>spamassassins autowhitelist or bayes database currently don't work with
>mimedefang, because they don't lock the database so concurrent access will
>destroy them.

Actually, SA *does* lock the Bayes database.  The problem is that when it 
needs to expire old entries, it takes a *lot* of memory.  Since it's being 
called from MIMEDefang, if you've placed a limit on the amount of memory 
allowed per MD slave, it can easily hit that limit, the slave will be cut 
off, the lock file will stay in place, and no further MD slaves will be 
able to access it.

In other words, the access problems are the result of a stale lock file, 
not concurrent access.

Check /etc/sysconfig/mimedefang and see if you have MX_MAX_AS defined.  Try 
removing it or increasing it.

When I first activated Bayes on my setup, we had this problem so frequently 
that I wrote a cron job to check for errors in the log and automatically 
restart the multiplexor and delete the lock file.  Someone pointed out the 
memory issue, so I upped MX_MAX_AS to 120000 on April 14, and there haven't 
been any problems since.  It had been hitting at least daily before, and 
since then it hasn't even been triggered *once*.


Kelson Vibber
SpeedGate Communications <www.speed.net>