On Thu, 2 Aug 2012 07:37:02 +0000 Daniel Lemke wrote: > We have a strange problem with our Bayes filter here, looks like the > learning and/or the journal sync regularly causes a high scan times > of about 90 seconds. It occurs once every day at nearly the same time > (around 8.04pm). > > This is an extract from spamd logs when the problem just occurred: > http://pastebin.com/iutSEajZ > > Debug log for the corresponding period of time can be found here: > http://download.jam-software.de/SaWin/spamdDebugLog-Jul26.txt > > We already tried disabling Bayes autolearn and set the > bayes_journal_max_size to 0, without success. This is kind of curious > as it occurs every day at the same time, no matter how often journal > or token sync was done before. So the token or journal file sizes > apparently don't matter.
If it weren't for its happening at the same, it would almost certainly be due to autoexpiry. The standard advice would be to turn-off autexpiry and expire by running sa-learn --force-expire from cron (or Windows equivalent). If you aren't already doing this it's worth ruling it out.
