Multiple Parent problems

Dear All,

   I am having trouble with using squid as part of a
highly-available cache set-up. The version I am using
is 1.1.20 with RETRY_CONNECT and US_ICMP defined.
My host is running Solaris 2.6, and there is an apache
server locally serving the cachemgr information.

   I have two parent proxy-servers (non-squid), and
I am trying to configure my squid proxy so that if one of
these servers goes down then the remaining parent
will be used automatically. I think I have located a
symptom of the problem...

   When looking at the cache server list from cachemgr
I can see the following entries...

Parent : lonss24318.uk.ml.com/80/7
Status : Up
AVG RTT : 0 msec
ACK DEFICIT: 0
PINGS SENT : 0
PINGS ACKED: 0 0%
FETCHES : 14 0%
IGNORED : 0 0%
Histogram of PINGS ACKED:
DOMAIN LIST:

Parent : lonss24317.uk.ml.com/80/7
Status : Up
AVG RTT : 0 msec
ACK DEFICIT: 0
PINGS SENT : 0
PINGS ACKED: 0 0%
FETCHES : 15 0%
IGNORED : 0 0%
Histogram of PINGS ACKED:
DOMAIN LIST:

However, since lonss24318 is switched off and unplugged
I do believe that its fetches entry should be 0 and its status
should be "down". By playing with the cache-conf file I did
manage to get them both to be labelled "down".

   Things I have tried that fail when one server is off...

   Configuring both parents in squid.conf, no-query and round-robin
        - half the items on a page download fast, the other half never
arrive.

   Configuring a virtual parent with two IP addresses mapped.
        - the duff IP address never seems to be timed out.

   Mike Pelletier clued me in to look for a "cache-parent-retry" patch,
but I've not been able to locate it (it sounds just right).

   The ability to elegantly handle the failure of a parent is very
important to the project I'm working on. Help would be appreciated.

   At the moment I'm considering a shell script that will ping both
parents, choosing a configuration file and kill -HUP'ing the squid
if the configuration needs to change. Thats icky though.

   Tim

   P.S. the use of port 7 in the ICP field, and web server
acceleration are also areas that I am finding a bit opaque,
but I don't think that these are causing the above problem.
Received on Mon Jan 26 1998 - 08:17:25 MST