Re: [squid-users] preventing 'DEAD parents'

Torsten.Lange@GECITS-EU.COM wrote:
> i sometimes have squid processes that report
> 'detected DEAD parent', although the parent is reachable.

Do you also get "TCP connection to parent failed" or a similar message,
or only the "detected DEAD parent" message?

If only the DEAD message, then try adjusting icp_timeout. It can help.

> is it possible to disable the timeouts for child <-> parent
> communication and thus, stopping squid from complaining about
> dead parents?

Well.. there must be a timeout, but sometimes if the latency varies or
differs too much between your peers then the automatic timeout selection
can be a bit to aggressive. In such cases setting icp_timeout can help.

> what exactly does the 'default' tag mean for the cache_peer parameter?
> i understood that i should use a 'default' parent as a last-resort parent.

Unless it has been declared dead.

Selection priorities is roughly

digest hit
ICP first hit
ICP first miss
default
round-robin
any parent

Note: in most child->parent relations I recommend running without ICP
(the no-query cache_peer option), and instead use the cache_peer
connect-timeout option to quickly detect dead peers.

--
Henrik Nordstrom
Squid Hacker