I wonder why dual WAN loadbalancing/failover is so hard for Linux in general. I've been trying to get a Vyatta VC4 install up and running, but run into a problem where a combination of load balancing and NAT breaks things. My boss finally pulled the plug and ordered a hardware solution -- too bad that, as Vyatta VC4 is Lenny based. Given some time, I could probably figure out where it was failing. As far as I can see, the Vyatta system could tell that one of the WAN connections was down and adjust it's internal routes successfully. But it left the kernel routing table in place. So if the kernel routing table said:

Destination Gateway (snip) Iface
0.0.0.0 10.1.0.1 eth1
0.0.0.0 10.2.0.1 eth2

and eth1 dropped, the Vyatta box would simply stop talking.

Probably just a simple hook somewhere in Vyatta's internals to specifically set and unset the kernel routes when it detects a routing change.

But -- now I've been told to move on. :(