On 27 May 2010, at 5:09, Chris Sarginson wrote: >> Externally, users can only hit our sites at www.example.com or >> someapp.example.com, but internally for testing, monitoring, and >> configuration purposes, we want to be able to hit the backend servers >> individually (i.e. web[1-3].www.example.com, web[1-3].someapp.example.com). >> I could make 3 separate backends to handle each of these and use a >> use_backend acl to pick based on the Host header, but that seems like >> overkill and would be the only purpose those backends serve. Is there a way >> to balance to a _specific_ server based on the Host header (no cookies)? I >> know you can use a balance hdr, but that is doing a hash of the header to >> pick the server. I need to explicitly configure `Host: >> webX.___.example.com' to go to the backend server webX. Is that doable >> somehow, or do I have to go the separate backend route? > > I find the easiest way to do this is to add an alias to each host like you > mention, and then set up DNS records to point to that servers external IP > address, bypassing haproxy. This way if you do start experiencing any issues > you aren't having to debug 2 layers to find the cause - it's either the host > or the loadbalancer.
I had considered this, but there are two issues with it: 1) If we are testing something, the stack now looks different (we are not going through the load balancer), and 2) I don't want to set up SSL on the servers (it is all handled by stunnel on the haproxy box). So, I still want all of the traffic to go through haproxy. Based on the lack of other ideas, I guess I will have to go the separate-backend-per-server route. :-\ Thanks!
