Just incase I've been giving wrong impression, I actually got LVS DR working for almost 24 hrs already. I had to turn off drop_entry, drop_packet, and use amemthresh 1024 due to memory load of the MySQL NDB cluster storage engines.
LVS DR routing isn't setup properly by ha-lvs. There is no lo alias
interface, yet there is routing. So far that's the only gotcha I see.
Packets destined for the supposed lo alias on the realserver aren't
responded to by the node.