Skip to content

Raspberry Pi k3s fleet — network and power prerequisites

Parent runbook: How to provision k3s, Longhorn, and Rancher on a Raspberry Pi fleet.


Network — mandatory

  1. Stable L2/L3 connectivity between all cluster nodes (same site for P0/P1 unless you are explicitly building multi-site later).
  2. Reserved IPs or DHCP reservations; document hostname or IP used in K3S_URL.
  3. DNS or /etc/hosts consistency for TLS names (Rancher, ingress, Longhorn UI if exposed).
  4. Firewall rules documented per k3s networking requirements at install time.
  5. Time sync available (internet NTP or local stratum server).

Optional (HA / scale)

  • Secondary uplink or out-of-band way to power-cycle remote Pis (no IPMI on Pi—plan smart PDU or on-site access).
  • Site-to-site VPN only when multi-site etcd or storage replication is explicitly architected (later).

Power — mandatory for P1+

Requirement Why
Clean shutdown budget etcd and Longhorn dislike unclean power loss during writes.
Known brownout behavior UPS or generator discipline at farm edge—Off-grid power strategy — Demory.
Labeled circuits Partial cluster power loss produces confusing storage quorum symptoms.

Optional (HA / scale)

  • UPS feeding switch and core Pis together so the network does not die while nodes stay up.

Farm / off-grid coupling (context)

mermaid flowchart LR subgraph util [Utility] Grid[Grid / inverter] Batt[Battery] end subgraph net [Cluster LAN] K[k3s nodes] Sw[Switch] end Grid --> Batt Batt --> K Grid --> Sw

If optional farm power (Popt) cannot sustain cluster plus network baseline load, do not claim a tight P1 RTO: run fewer stateful sets or shut down gracefully before the battery floor. Align with Off-grid operational decision rules — power and networking (Demory) when those constraints apply.