RE: ICP Port

Hi Henrik,

We recently had a problem where some of our parent cache's lost their
network connectivity to the outside world (connectivity from us to the
parent cache's were fine).

Our parents Squid was returning network unreachable messages after a
considerable delay.

My question is, how can Squid be changed to better handle this type of
failure? I was thinking if I enable something like test_reachability, if the
site is not reachable by that parent cache then Squid will use another
avenue?

Your thoughts?

Thanks,

Chris.

-----Original Message-----
From: hno@hem.passagen.se [mailto:hno@hem.passagen.se]
Sent: Friday, July 17, 1998 6:37 AM
To: cgalanos@cleanweb.net
Cc: squid-users@ircache.net
Subject: Re: ICP Port

Chris Galanos wrote:

> I don't think I am totally understanding the concept of ICP.
> Obviously I want it to make UDP requests to both of the parents to
> see which one is fastest, yet I want it to rely on one of the parents
> if the other goes down. In other worlds say I have:
>
> cache_host cache.domain.com parent 8080 0

This setting (ICP port 0) does not work unless you use
single_parend_bypass on
or add the no-query keyword (which does not work reliably in 1.1.X
possibly including 1.1.22)

If you want Squid to check if the parents are alive then the parents
need to have at least simple TCP services installed (UDP echo), and you
can then configure the parents in Squid to use ICP port 7 (UDP echo).

If the parents support ICP then configure Squid with their ICP ports.

The key issue is that you can't use UDP queries unless there is someone
who listens to them. To support a approximation of ICP on non-ICP
servers Squid supports the use of the UCP echo port to measure if the
machine is alive and connected.

---
Henrik Nordström
Sparetime Squid Hacker