[Linux-cluster] LVS/Piranha Failover Problems

[David Brieck Jr.]

We have been using Piranha for about 7 months now and we've
consistently had a problem failing over. All of our systems are 4.4,
but we're planning 4.5 and 5 in the near future. We have two load
balancers, one is supposed to fail over to the other. Behind the load
balancers we have 3 real servers. The load balancer has 12 ip
addresses that it uses to send traffic over to the real servers. Each
real server is configured the same way.

The problem we have is whenever we fail over the master LB to the
slave, one of the real servers refuses to use the fail over as it's
gateway causing outages for some people. It's configured exactly as it
should be, which is just like all the others but it still refuses to
function.

Another problem we have is that not all of the virtual server's public
IP addresses are loaded when it fails over and even on a restart of
pulse. Everything would indicate they are loaded by pulse, but they
don't show up in 'ip addr list'. If I manually run the ifconfig line
for each missing public IP everything works properly. I was originally
told this problem had to do with the channel bonding we were using,
however we have since stopped using it, but the problem persists.

Any help would be greatly appreciated.

David