John Hardin wrote:
On Mon, 15 Nov 2010, Kris Deugau wrote:
I noticed recently that the average ~0.8s scan time on our filter
cluster had jumped to just over 3s.
If you can duplicate this behavior on a manual scan of a test message,
there are debug flags and a timing plugin that will help troubleshoot
performance problems.
Thanks! It *was* reproducible with a given message, although it took
re-reading the debug output several times (after a fair bit of "why is
my debug instance not loading most of the stock rules?") to find the
problem:
I had forgotten to restart one rbldnsd instance after moving the local
URI DNSBL from a tinydns data fragment into rbldnsd.
And that rbldnsd instance is the one used by the first resolver in
/etc/resolv.conf on both spamd hosts...
*headdesk*
(For the archives: If the HitFreqsRuleTiming.pm plugin doesn't report a
regular rule occupying any significant chunk of the scan time, you've
probably got a DNS issue somewhere.)
-kgd
