changes:
core:
- cache query hung if it tried to read from
a closed socket (in case of a manual cache kill).
NOTES:
I've today made some stresstests again.
I "bombarded" our mailserver with 50 connections and sending
1000 mails per connection. They all were designed to be blocked.
The cache had an efficiency of 15.100 Mails/minute (21.744.000 mails/day)
While under fire I did some evil stuff to the childs, killing them,
killing the cache and so on. No deadlocks.
Normal mails sent while under fire arrived instant, too.
Test was made on a P3 dual, FreeBSD.
graph: http://www.ek-muc.de/mailgraph.cgi
stats: http://www.policyd-weight.org/stats.html
Note the 30-days calculated bandwidth saving: 53GB
Also note, that the CPU usage is still below postfix.
--
Robert Felber (PGP: 896CF30B)
Munich, Germany
____________________________________________________________
Policyd-weight Mailinglist - http://www.policyd-weight.org/
