I've been having a bit of trouble getting ipvs working on 2.6 kernels.
I'm using keepalived to manage a cluster of web/application servers from
a pair of load balancers, using vrrp to handle router failover. The
setup is a basic weight-round-robin LVS-NAT setup. I have it working
(seemingly) fine on one load balancer running a 2.4 kernel, but on the
other one, using 2.6.10, 2.6.9, or 2.6.8 (I've been trying it off and on
for a while) I can start up keepalived to set up the cluster, but then
if I run ipvsadm it hangs. The ipvsadm process then becomes an
unkillable zombie. While this does raise interesting possibilities
about creating unstoppable armies of zombies, in practice I have to
reboot the server to get it usable again.
Has anyone seen this ipvsadm lockup before? Any theories as to what
might be causing it?
Thanks,
Rob
|