When metallb sets listeners, it does that in a loop and if one fails, then the whole process is aborted

Always, if we manage to make setting the ndp responder fail on a ipv4 cluster

This was raised upstream. 

1. IPv4 only cluster 
2. Interface with ipv6 addresses
3. MetalLB fails to set the ndp responder. What I saw in the upstream issue was that metallb failed to bind to the interface to listen. This caused a cascade effect that prevented l2 to work even on ipv4

MetalLB L2 not working with ipv4  

MetalLB should be more resilient. Not being able to listen on one single interface should not prevent it to work on the others.