You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@spamassassin.apache.org by Ronan <r....@qub.ac.uk> on 2005/10/21 10:47:34 UTC

round robin

In a heavily loaded server with a lot of connections hitting it per 
second, what is the best way to run spamd...

the standard 3.1 (hot children) or the old style round robin???

I have been having problems over the last 2 weeks etc with the spamd 
timing out connections from the mta's i run so i switched to round robin 
and i havent had an acl error since... is there a way to increase the 
number of hot children from the default??

Anyone care to comment on this?

R

Re: round robin

Posted by Matt Kettler <mk...@comcast.net>.
At 04:47 AM 10/21/2005, Ronan wrote:
>In a heavily loaded server with a lot of connections hitting it per 
>second, what is the best way to run spamd...
>
>the standard 3.1 (hot children) or the old style round robin???
>
>I have been having problems over the last 2 weeks etc with the spamd 
>timing out connections from the mta's i run so i switched to round robin 
>and i havent had an acl error since... is there a way to increase the 
>number of hot children from the default??


I think the number of "hot" children is a function of the total number of 
children.

Really, using the hot child method should increase overall performance, not 
lower it. If there isn't a "hot" child, then the master spamd should send 
the request to a non-hot child for processing without delay.  All the "hot 
child" algorithm should do is favor a subset of the children when there's 
low-load. However under high load it should utilize all children, not drop 
connections.

It sounds like there's one of two situations:

1) a bug in the hot-child scheduler (ie: spamd won't schedule to the 
non-hot children, effectively acting as a round-robin only over the hot 
children and ignoring the others)

2) your problem wasn't really fixed by the switch to round robin, it's just 
a function of load levels that you've not seen since you restarted. In this 
case you've just got too few children available in general.