Skip to content

Raspberry Pi k3s fleet — validation checklist

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

Cluster — mandatory

  • kubectl get nodes: all nodes Ready without repeated NotReady flapping.
  • kubectl get pods -A: core add-ons healthy (resolve CrashLoopBackOff before sign-off).
  • Time sync verified on each node (logs or timedatectl).
  • Disk headroom on the etcd server node and on every Longhorn disk.

Longhorn — mandatory when Longhorn is in scope

  • Disks schedulable on expected nodes (Longhorn UI or kubectl).
  • Test PVC: bind, write, read, delete cycle succeeds.
  • Replica policy documented per StorageClass or volume; observe CPU during rebuild (simulate with cordon/drain in lab).

Rancher — only if installed

  • HTTPS endpoint loads with trusted certs (or documented lab exception).
  • Admin bootstrap password rotated; break-glass access documented.
  • Helm chart/value version pinned; upgrade path noted.

Backup — P1+ mandatory

  • Application logical backup (e.g. pg_dump path) ran successfully at least once; artifact exists off-node.
  • Longhorn backup job succeeded, or deferral is explicitly logged with risk acceptance.
  • Tabletop restore completed against a non-production namespace or lab cluster.

Performance sanity (Pi — mandatory honesty)

  • Idle and loaded power measured for server + one agent (watts at wall where possible).
  • During Longhorn rebuild, no sustained 100% CPU on all cores for routine work (kubectl top nodes if metrics exist, or htop on node).