Re: Last resort for dead parent

Niall Young wrote:

> dead_peer_timeout depends on ICP replies to mark a parent cache as dead
> right? What happens if we're not using ICP at all - we'll never mark
> this single parent as dead so if it goes down we're screwed. Is there
> any way around this?

Squid also counts TCP connection failures when deciding if a peer is
dead or alive. In Squid 2.2 this detection is more or less transparent
(except for the delay for timeout if the parent server is fully gone),
but Squid <2.2 may show an error message to the client until it decides
that the parent is down.

I also have some patches to Squid 2.2 to improve detection of dead
non-ICP peers. See http://hem.passagen.se/hno/squid/

--
Henrik Nordstrom
Spare time Squid hacker