Skip to content

Raspberry Pi fleet provisioning standard — smart farm / homelab

Purpose

Minimum operator standard for Pi-class nodes joining a k3s fleet that may run farmOS, telemetry, and Longhorn—aligned with official k3s install notes and Longhorn on K3s prerequisites.

Standards catalog: Operational standards — farm and homelab platform.

Doctrine package: Platform doctrine package — homelab / farm edge.

Sources: k3s quick-start, Longhorn CSI on K3s, Pi + k3s + Longhorn homelab narrative (secondary). External: K3s requirements.

Checklist (every node)

Item Why
Unique hostname k3s quick-start: duplicate hostnames break joins—use K3S_NODE_NAME if needed (quick-start)
64-bit OS Match k3s + container images for your stack; verify against requirements
Quality storage for etcd / Longhorn SD-only control plane is a fragility pattern; USB3 SSD or better for server nodes and Longhorn disks per homelab narrative
open-iscsi / iscsiadm Longhorn on K3s prerequisite (CSI capture)
Time sync NTP/chrony—cluster security and logs correlate to time / PNT topics
Network policy known Static DHCP or reserved IPs; document server URL and token handling per quick-start

Roles (k3s)

  • Server: runs datastore + control plane (embedded or HA topology per architecture).
  • Agent: workload + kubelet; join with K3S_URL + K3S_TOKEN (quick-start).

Longhorn scheduling: narrative suggests disabling Longhorn replicas on server nodes if you want control-plane separation—validate against your HA and disk layout (homelab capture, Longhorn docs on node/disk eviction).

Operational boundaries (this repo)

Boundary Rule
Node count Phase 0/1: one k3s server + agents unless platform decision memo gates justify HA.
Disk No production farmOS without SSD-class storage for etcd / Longhorn paths on server nodes—SD-only is fragile (pilot only).
Network Stable LAN and reserved IPs; edge farm nodes must match power and WAN plans (Two-site operations).
Secrets kubeconfig and join tokens are credentials—not checked into git (remote access posture).

Pilot vs production-ready

Aspect Pilot (acceptable) Production-ready (raise bar)
Control plane One k3s server HA servers per k3s docs + tested restore
Storage Single replica / small PVCs Longhorn replica count + backup target + monitored disk health
Secrets Manual token files Sealed Secrets / external secret store (policy choice—not prescribed here)