There are two primary ways to configure Squid to be a load balancer.
Private /etc/hosts
The easiest is to simply list multiple ips for the httpd_accel_host in
your /etc/hosts file and Squid will automatically round-robin across
the backend servers.
Use cache_peer
This is more complex. Squid has advanced support for communicating with
other caches. It just so happens that this communication happens over HTTP
so you can set Squid up to treat the web servers you wish to load balance
as peer caches. The configuration would look something like this:
httpd_accel_host www.visible-domain.com
httpd_accel_uses_host_header on
never_direct allow all
cache_peer server1 parent 80 0 no-query round-robin
cache_peer server2 parent 80 0 no-query round-robin
cache_peer server3 parent 80 0 no-query round-robin
no-query in the above tells Squid not to try to send ICP (Internet Cache Protocol)
requests to the Apache servers. You could turn on the echo port on each server and redirect
these ICP requests to there which would make this system automatically detect a server that is
down.
Avoiding Redirectors
It is generally a good idea to avoid external redirectors. A lot of things can be
done directly in Squid's config file. For example:
acl domain1 dstdomain www.domain1.com
acl domain2 dstdomain www.domain2.com
acl domain3 dstdomain www.domain3.com
cache_peer_access server1 allow domain1
cache_peer_access server2 allow domain2
cache_peer_access server3 allow domain3
cache_peer_access server1 deny all
cache_peer_access server2 deny all
cache_peer_access server3 deny all
This would configure Squid to send requests for pages on certain domains to
certain backend servers. These backend servers could actually be aliases to different
ips on the same server if you wanted to run multiple Apache instances on different ports
on the same box.
