[bitfolk] Scaling spamassassin bayes db

Hello, In the last month I've been looking at improving the performance of the spamassassin spamd service. I added two more hosts to the load balancer and that has helped some, but the bottleneck has now moved to the bayes database. The servers are connecting to a MySQL bayes database so that they're all in sync, but the problem comes when the database slows down. Queries stack up and spamd doesn't seem to have config for giving up on bayes after (say) 10 seconds, so then those connections back up until all the servers are full. Normally I'd look at scaling it by doing multi-master replication, but the spamassassin docs specifically say this isn't supported and won't work. Has anyone scaled a spamassassin bayes DB before? Is my only option to replicate it to slaves with a MySQL proxy in front to send writes to the master? Maybe I could revert to using DB_File databases, but over NFS. Except it sounds horrible. Cheers, Andy -- http://bitfolk.com/ -- No-nonsense VPS hosting