On 11/29/07 2:23 AM, "[EMAIL PROTECTED]" <[EMAIL PROTECTED]> wrote:
> But to get hold of the timeout problems and the bad performing bayes db we > did what the spamassassin people suggest since quite some time: > > - use global bayes instead bayes per user > - do not use auto_expire in local.cf > - run instead a cronjob once a day that expires old tokens This is what I have configured, except with DBM instead of SQL (next on my agenda), and expire every 3 hours to keep the DB from getting out of hand. > Result: > - basically never more that 300k tokens in the mysql bayes table. > - cronjob expires 150k - 250k tokens per day. One might think how old the > tokens really become... > - good spamd scanning performance. I'm expiring about 3.5 million tokens every 3 hours... > Manual learning is handled no differently than auto-learning. If this is true, then that tells me that with our volume, we either have to do all automatic learning, or all manual learning. With both enabled, any manual learning would likely be lost within less than a day. Ugh. Wes
