Ultra Monkey HC,HA,LB (LVS, HA-Linux), Layer3 Switch, No-Go?
 

I have four servers in a configuration similar to http://www.ultramonkey.org/3/topolog...-ha-lb-eg.html set up, and I believe all is set right. However, when I go to the virtual IP instead of getting the site I expect to see I eventually get a "Cannot find server" (IE) or a neverending "Connecting..." with FireFox. I disabled all firewalls to make sure they weren't the problem, and it was still a no-go. It seems to me like this is what should happen, as requests to the virtual IP are directed to the master server by the switch, and they aren't sent anywhere else. I think that theory is backed up by the fact that arptables prevents the virtual IP from being advertised on the actual servers. My question is, is that theory correct, or is something wrong with my configuration? I will admit that I haven't fully researched the way this configuration is meant to function, but my theory makes sense to me, and I don't understand how it could possibly be advantageous to run this configuration if I had to substitute hubbing for switching. Assuming that my theory is correct, is there anything I can do about it (Cisco 3750G)? Assuming that it is possible to run this configuration, what might I be doing wrong? (arptables -nLv, ip addr sh eth0, ipvsadm -nL, /etc/ha.d/resource.d/LVSSyncDaemonSwap master status, and ip addr sh lo0 all show the expected output on their relative servers). Any help appreciated.

update
 

Well, I kept reading and now I understand why my theory is wrong. However, I still don't understand why the site I have configured isn't coming up. I tested failover between Linux Directors and Real Serers as well. Everything functioned as expected except the site doesn't load. The site loads fine when I point directly at one of the Real Servers by name or IP. I'm going to shout down iptables again and scour the config again, but I'm getting nowhere thus far.

Hi,

we have had a problem (but this was for a pool of DNS servers instead of a pool of web server).
Have you configured the http server so that the IP address for the answer (route back to the client) is the routeur?

Hey Vincent,
Actually, I got it working. This was some time ago & I don't remebber how. Those server have since been reused for something else,and I don't currently have anything runnig LVS. However, I had a couple other problems I don't think sI ever posted anythere. One was that I couldn't make one LVS server primary for one cluster & another LVS server primary for another cluster, (so one server handeled all the load & the other was only standby). I should think that would be doable, but I don't know, and I can't provide any configs to exemplify my attempts that didn't lead to succes. I would like to know, though, whether or not I should be able to do that. It might be important to note that restarting the servers 0ead to perfect failover every time, but once the primary server kernel panicked, and the other didn't take over, ever again (I didn't really do much to troubleshoot this, since it was testing & I had begun to fucos on other things).
Dustin