Ok. Now I see your problem.
Timeouts won't help you much. Your problem is that the return traffic to
parent#1 is lost.
Since "parent#1" is still reachable by "sibling" the dead-peer detection
won't go in.
To make it work "parent#1" must either be able to use different outgoing
addresses depending on if the link is up or not (i.e. if it contacts
"parent" or goes direct), or it has to return a new errorcode to
"sibling" causing it to go direct.
You may be able to solve the problem at "parent#1" by playing around a
bit with the routing instead of using tcp_outgoing_address, or by
extending the cache_peer directive with a outgoing_address to use when
contacting the sattelite peer "parent".
Squid is not designed to handle this specific case where a parents
internet connectivity goes dead.
/Henrik
Paul wrote:
>
> > Not sure I entirely understands your setup. What do you mean by parent
> > to parent problem?
>
> parent --Satellite---> parent#1 <---> sibling
> | | |
> ---<------Net-------<-----------------
Received on Thu Feb 03 2000 - 14:50:55 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:50:56 MST