Scale-out DHCP behind a VIP.

Add a node, keep the same pools. Relays stay pointed at one address.

Each lattice-dhcpd process is identical. Lease state lives in Postgres, not in a failover peer database on the box. You scale by adding processes and keeping the VIP.

Listen

Production: UDP :67 on the VIP, typically via relays. Lab and compose: :6767 so you can run without capabilities on a shared host.

When a license is over limit

Existing leases still renew. New Discover/Offer allocations and new reservations are refused. DHCP does not crash to enforce a cap.