With the OVS mechanism driver, it's allowed to have multiple segments on the same compute for the same physnet. It's been developed and merged last year. 

Customer requests the support for this feature with ML2-OVN. 

The unavailability of this feature forces the cloud operators to create a new network whenever one of the subnets is full (for a given physnet). Then, users have to change their pipelines, IaC, and configuration to use the new network with the free IP addresses in the created subnets. Users cannot stack subnets in the same network which increases complexity and the management of the OpenStack platform.

Following is the corresponding spec for ML2-OVS upstream

Acceptance criteria:

1. A single host can successfully host VMs on multiple different segments of the same routed provider network. VMs must be able to communicate with their respective gateways and any external networks.
   2. The underlying OVN and OVS configurations are correctly applied on the compute host to handle traffic from multiple segments.
   3.Neutron DHCP agent (or OVN's native DHCP) can service requests from multiple segments for the same network on a single host.
   4. verify that Nova scheduler is aware that a host is a valid candidate for all segments it is mapped to, not just the first one.
   5. VMs on different routed segments cannot communicate at Layer 2, as this is a routed (L3) network.

Definition of Done:

1) Document all the limitations encountered while implementing/testing this feature
2) Document "config guide" for this feature (along with topology if possible)
3) All QE test should have been automated, executed and passed in CI
4) Prepare material/doc to enable Support team on possible customer issues

https://review.opendev.org/c/openstack/neutron-specs/+/823823