No, we did something a little different. We have some NFS shares to some tmp directories on the machine. We copy the compressed bayes files over there and then dump a control file to it. There is a cronjob set to run every x minutes that looks for the control file. If found then we untar then into a new directory /etc/mail/spamassassin/bayes_new and then we stop postfix then spamd, rename the old directory and then start spamd then postfix back up. We have an entry in /etc/init.d/spamd set to chmod and chown the directory and files to the proper user account. When completed we remove the tar file and the control file.
The script them archives the old folders. We keep /etc/mail/spamassassin/bayes.1 - bayes.9 as backups. You never know. We have it set to do this ever 2 hours and have it set to different times on different nodes of the cluster. The files are dumped every 30 minutes to all servers in any event. That is the gist of it. There is a little more going on than that which is specific to our configuration (mostly postfix and not SA related). For you newbies out there. If you use something similar you should note that you do not want a persistent connection to your NFS on your other servers. Only connect when you need to. We took down one of the nodes for a test and forgot to unmound the connection. This caused the master node to run horribly. >From what I understand I think the long term solution for the bayes issue is solved for us in 3.0 though. (OR DID I MISS UNDERSTAND THAT IT WOULD HAVE SQL SUPPORT?) We will try it with a mysql backend server and then dump bayes to it. This way there is no master node and each machine can add it's own entries. We can then replicate the database to our offsite location on a very regular basis (or real time replication). -----Original Message----- From: D.J. [mailto:[EMAIL PROTECTED] Sent: Friday, July 16, 2004 8:42 AM To: [EMAIL PROTECTED] Subject: Re: Bayes Bit Me On Thu, 15 Jul 2004 10:17:54 -0700, Gary Smith <[EMAIL PROTECTED]> wrote: > It's common... The short answer is yes, copy bayes_* over, and restart. > We have a job that does this every so often to our offsite cluster. > > Actually, we do this on 6 machines. We take the bayes db files one of > our load balanced nodes and copy it to a bunch of different servers. We > have a single one that we use as a primary so we can manually feed it > daily. > > Gary Wayne Smith Glad to see I'm not the only one doing this :-) Are you using rsync to do this as I am, or something else? - D.J.
