There is no ISC failover protocol. Shared Postgres is HA for dhcpd.
Add a dhcpd node
- Host with chrony. Offset < 500ms vs the writer.
/etc/lattice/lattice.envwith a newLATTICE_NODE_NAME.- Enable
lattice-dhcpd.curl -sf http://<node>:8080/readyz. - Add to the UDP VIP backend. Relays stay on the VIP.
- Confirm a
noderow with recentlast_seen_at.
Drain
- Remove from VIP / withdraw anycast.
- Wait until heartbeat is stale and offer/ack logs stop.
- Stop the unit. Do not delete leases. Do not
DELETE FROM nodeunless retiring the name.
Postgres failover
- Fence the old writer (Patroni) or wait for the managed endpoint to flip.
- Processes reconnect.
/readyzis 503 until ping works. - Check
lattice_outbox_lag. Pending rows retry.
More: Clock · Pool exhaustion · Outbox · Backup.