Description of problem: With centralized dhcp only vms in the same segment that the controller obtain ips. Maybe a RFE should be opened.
Marking as RFE indeed. Routed provider networks were designed with distributed DHCP/metadata agents in mind (and in current requirements), this is also in reference architecture. Though this could be useful for some use-cases, I think most of the deployments will work fine with added agents in compute nodes (or if needed a separate networker node)