On Sat, Dec 08, 2012 at 12:13:15PM -0800, John Hardin wrote: > Are you willing to enable the HitFreqsRuleTiming and see if a > particular rule stands out?
It's done now. It actually ran much faster than yesterday: 10:23 hours for 10,190 messages or 3.668 seconds per message, with JOBS=4 and CPU load always near zero. This is on a residential ADSL connection, using the provider's caching name servers. Here is the top ten from timing.log: v1 T RAZOR2_CF_RANGE_E8_51_100 1.5125 1.5125 1 T PYZOR_CHECK 0.0717 0.0717 1 T __FILL_THIS_FORM_SHORT2 0.0119 0.0119 1 T __FILL_THIS_FORM_LONG2 0.0118 0.0118 1 T __FILL_THIS_FORM_FRAUD_PHISH1 0.0046 0.0046 1 T __FILL_THIS_FORM_LOAN1 0.0034 0.0034 1 T __FILL_THIS_FORM_SHORT1 0.0011 0.0011 1 T T_FR_DOT_FEVER_5 0.0011 0.0011 1 T USER_IN_DKIM_WHITELIST 0.0010 0.0010 1 T __FILL_THIS_FORM_LONG1 0.0010 0.0010 1 Full log is at <https://www.sedacon.com/private/timing.log.bz2>, in case you want to have a look. If those numbers are seconds, they don't add up: awk '{t+=$3}END{print t}' <timing.log returns 1.7167, not 14.672, nor 3.668. In any case, if the last number in that file is meant to be $mailsa->{rule_timing}{runs}, something seems to be wrong with HitFreqsRuleTiming... Regards Marc
