Marking proxy servers as zombie - odd behaviour

[Bjørn Mork]

Alan DeKok <aland at deployingradius.com> writes:
> John Horne wrote:
>> Hmm. Given that the servers are lightly loaded, I guess we are looking
>> at packet loss over the network? 
>
>   Yes.  Many packets lost.  The NAS re-transmits, FR re-transmits, and
> the home server doesn't respond.
>
>   The default timeout before marking a home server zombie is 30s.  If
> there aren't *any* responses to retransmits in 30s, something is very wrong.

Given these conditions, I believe it's much more likely that the home
servers *choose* not to respond to some of the requests.  There are lots
of reasons why this might happen.  Some are even default FreeRADIUS
behaviour (if no module handle an accounting request, then it isn't
acked).

I would start by looking for any such deliberately ignored request.  As
you say, packet loss is a very unlikely explanation.



Bjørn