> -----Original Message----- > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] > Sent: Tuesday, September 20, 2005 9:39 AM > To: Dean Baldwin > Cc: users@spamassassin.apache.org > Subject: Re: NFS and SA/Bayes DB > > -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA1 > > > Dean Baldwin writes: > > Hi, > > > > I currently have a cluster of inbound MXs running Amavis/SA etc. At the > > moment they are running as single entities in the sense that they do not > > share their databases. > > > > Has anybody tried this or any other method of shared db for multiple > MXs?
Yes, we did. It was brutal and bloody but it worked. Basically we had a master that we rsync'ed across the network to all of the endpoints on a scheduled basis (4 times a day). Then we had a job on each machine to swap those bayes DB's and restart SA. Did I mention it was bloody? > > It sounds like an SQL db is the best option. NFS sharing is certainly not > recommended. > > With any luck one of the other large scale users using shared SQL dbs > might pipe up with comments here ;) > We have 4 MX servers for incoming mail that call back to two SA servers. These SA servers use MySQL as the backend for both bayes and AWL. Seems to work just fine and actually seems to be a little faster for us as well. We don't use Amavis but instead use a commercial AV scanner that runs on it's own servers as well. Our backend MySQL is running Linux-HA with DRBD on a pair of Dell 4700's with 1gb ram. The SA servers are running on two Dell 4700's (with HT enabled) with 2gb ram. The front end MX servers are running on a mixed set of dell workstations. We have been playing with this environment for some time swapping hardware in and out but for the most part the environment has a very low load balance. Typical email is about 170k emails per day. HTH... Gary Smith