I've got a LVS-DR set up and responding on 1 director 2 realserver
setup (soon to be more after I resolve this issue).
I have resolved the ARP issue by putting the VIP on eth1 on the real
servers and no cable hooked up to eth1.
Obligatory network diagram
+-- [data feed]
+-- [real server 1]
+-- [real server 2]
+-- [other hardware]
What I'm seeing is that the scheduler for both rr and wrr seems to
"lock up" and fixate on one host for a while before apparently
randomly deciding to again distribute traffic evenly or weightedly.
Unless, I suppose, I didn't really resolve the ARP issue, but the VIP
is on eth1, with no cables on the real server and eth0:1 on the
When the scheduler is in "lock up mode", the stats from
ipvsadm -Ln --stats -t $VIP:80
All machines run kernel 2.4.18 upgraded RH. Yes, I know, the mantra
is to build from the latest kernel and IPVS patches, and I'm headed
that way now, but given that it *does* work intermittently, I thought
I might have done something or this might be a known issue that would
make it a shorter trip.
I've read the docs and tried several google queries all to no avail.