On 2008-05-12, Henrik Nordstrom <henrik@xxxxxxxxxxxxxxxxxxx> wrote: > On mån, 2008-05-12 at 13:52 +0200, Jan-Frode Myklebust wrote: > >> Thanks for the pointer, but that's not what we're trying to do no. It's >> a webserver cluster, with shared filesystem (GPFS), and heartbeath >> (linux-ha.org) for failover ip-addresses which are distributed trough >> rr-dns. So, no squid on our side, but the users might be going trough >> proxies, so it's important to test that everything works there too. > > If you have all that, why does it at all matter which server the request > got sent to? No, I agree .. it shouldn't matter at all, but when I don't have much controle over the application it's nice to verify that it works both in theory *and* in practice :-) Otherwise the customers might get unhappy, and we might get forced to use a system with sticky sessions towards each server. IMHO, seeing that this works, we now have a very nice scalable system, without too much complexity (simple failover ip-address)-- and also no single points of failure. -jf