There is no ISC failover protocol. Community is one node. Paid HA is three Ubuntu VMs in one site (same L2 for the VIP only) and a VIP (HA). Client VLANs do not share that L2. Extra site: --mode dhcp-only. Active/passive DC is restore onto a new cluster, then cut relays — not a stretched trio. HA — Active / passive DC.

Add a dhcpd node (paid)

lattice-install --mode join-token
lattice-install --mode dhcp-only --peer <primary-ip> --token <join-token> --ca-hash <hash>

Relays at that site point at this VM. Confirm a node row with recent last_seen_at. Join a control-plane member with ha-join, not dhcp-only.

Drain / upgrade (paid HA)

lattice-install --mode upgrade on a replica, then on the leader (switchover runs automatically). --force if switchover stalls. Do not delete leases.

Postgres failover (paid HA)

Patroni holds the writer. Processes use the VIP DSN. /readyz is 503 until ping works. Check lattice_outbox_lag. Pending rows retry.

Diag: lattice diag -o /tmp/lattice.tgz. Kubernetes examples (not the production SKU): k8s.

More: Clock · Pool exhaustion · Outbox · Backup.