Re: [squid-users] How to trick splay trees?

From: Amos Jeffries <>
Date: Tue, 31 Jul 2012 08:25:03 +1200

On 31/07/2012 1:25 a.m., Jannis Kafkoulas wrote:
> Hi,
> (I use squid 2.7. STABLE9 on RedHat EL 5.6)
> Following problem:
> I have following dstdomains defined
> going to par-std and par-alt cache_peers respectively:
> acl alt dstdomain
> acl std dstdomain
> Now I'd like "" to also go via cache_peer par-alt.
> Following two tries didn't work:
> # acl alt-2 dstdom_regex -i
> # acl alt dstdomain

The dstdomain one is faster. Both are correct for your requested policy.
The key word you stated being "also" ...

> The requests were sent to par-std cache_peer
> cache_peer parent 3128 0 no-query login=PASS proxy-only no-digest name=par-std
> cache_peer parent 80 0 no-query login=PASS proxy-only no-digest name=par-alt
> cache_peer_access par-alt allow alt-2
> cache_peer_access par-alt allow alt
> cache_peer_access par-std allow std
> Is there a way for that to work at all?

Unless given some specific selection algorithm (digest, ICP, hshes,
carp, roundrobin etc) Squid lists peers in configuration order when
attemping to pass traffic.

As I said above the key word in your policy statements is "also" - with
both peers *available* for use Squid will pick the first one that works.
With par-std being listed first your logs will show it being used until
such time as it becomes unresponsive or overloaded. Then par-alt will
pick up the slack for that one domain.

I think you are looking at the logs and seeing only par-std, thinking
its not working when actually it is. You can test by changing the order
of cache_peer definitions in your config and seeing the preferred peer
switch to the par-alt when the new ACL is added.

NOTE: you canot send a request via *both* using TCP unicast links, just one.

Received on Mon Jul 30 2012 - 20:25:15 MDT

This archive was generated by hypermail 2.2.0 : Tue Jul 31 2012 - 12:00:02 MDT