Thanks for the quick response Mike. Which option is it for passing the clients source IP? I haven't looked to configure anything like that.
I realised my test harness was on both networks, i've now run the tests again with it just on the 10.x network while failing haproxy over a couple of times. I'm only getting a handful (out of 100 users) socket resets every time I fail it over. Looking at the routing table the source IP must be the haproxy servers as i'm unable to see the 10.x network from the backend servers. So haproxy is handling the whole request. Make sense? 2009/11/4 Michael Miller <"mm-haproxy"@aquaorange.net>: > Hi Matt, > > Sounds pretty straight forward. You don't make it clear if the HAProxy > servers are passing the client's source IP to the backends or not... > that could affect your return path routing. If your backends route out > back through the HA proxies, you might see asymmetric routing. If your > HAProxy instances are infact working as proxies (i.e. they initiate > connections to the backends themselves) then your setup should work fine! > > Regards, > Mike > > Matt wrote: >> Hi, >> >> I'm currently looking at setting up HA proxy in an Active/Active >> cluster and have the following set-up that appears to be working. >> Does anyone have any thoughts on this? or where there could be an >> issue with client -> backend traffic? >> >> Two HA Proxy servers running in a NAT type configuration, that is eth0 >> on 10.x and eth1 on 192.x range. >> Traffic comes in on two virtual IPs (managed by heartbeat/pacemaker) >> that are on the 10.x range and the backend servers are all on the >> 192.x range. >> >> In the haproxy config I have the frontends listening to all interfaces >> with *:80 and everything is in httpmode. The config is the same on >> both servers, so both haproxy load balancers balance to all backends. >> >> Could this be an issue? >> >> Failover seems to be really simple as I don't need to restart/reload >> the haproxy service as haproxy is already listening on that interface. >> >> I've run tests against both haproxy servers while failing over the >> virtual IP, and while connections to the failed haproxy get reset at >> the point of failover, new connections work fine. >> >> I was worried that there maybe some routing issues with this setup, >> anyone else done similar? >> >> Thanks, >> >> Matt >> >> >
