Comment 8 for bug 1764738

We have a very similar use case.

In the future, I can see the need for wanting to have a compute node that can belong to 2 different routed networks.

However, Our current need/use case is that we want to have a compute node that is connected to multiple segments on the *same* routed network. Our network is an L3 spine+leaf design where the L2 is terminated at the ToR. The need for multiple segments to the same host is driven by our networking teams desire to keep the broadcast domains for the L2 small. Currently, each segment (vlan) has a /22 worth of ip address space, and a segment covers all servers attached to that ToR. The desire is to add additional vlans and thus subnets to the ToR and thus all servers under that ToR, to allow more VM's to be booted on the existing HV's. We need to have anywhere from 3-8 /22's of capacity depending on the number of hosts and the size of the HV's in the racks. You can imagine having ~8000 ip's in the same L2 vlan will start having broadcast domain issues. Hence the desire to break that up into smaller domains. In our clouds that are not yet running routed-networks, we do this without issue just by mapping in additional networks to the compute nodes.