Re: [squid-users] Squid-cache clustering

fre 2006-04-07 klockan 09:53 -0700 skrev Ben Drees:
> I am curious about a related question:
>
> In reverse proxy scenarios, what are the options for load balancing
> cache misses among several origin server replicas?

I would make each origin a peer to Squid, allowing Squid to use
round-robin etc for balancing the load.. Just remember to set the
connect timeout fairly short for better failover performance.

> 2) It is my understanding that if DNS returns more than one address for
> a hostname, Squid can be configured to perform round-robin selection of
> an origin server. Are there any caveats to be aware of when persistent
> connections are used between squid and the origin servers?

Yes, and it does by default. The problem is if/when there is a problem
with one server, it may then become very hard to make Squid use that
server again..

> 3) It seems like the Redirector API could be used as a hook to do this
> kind of load balancing also, offering a convenient place to code custom
> health checks.

I would avoid this if possible. Redirectors is quite resource demanding,
and it's tricky to get the setup correct when rewriting the URLs while
proxied... (cache gets split, backends may get confused about the actual
requested domain etc..)

Regards
Henrik