Skip to content

Kubernetes platform backup / DR — Pi fleet, k3s, Longhorn

Purpose

Integration page: maps parallel backup tracks (application logical, Longhorn volume/system, etcd, Rancher, secrets) to the k3s + Longhorn platform described in Platform doctrine package — homelab / farm edge. Sits on top of generic Backup strategy comparison — farmOS, homelab, PostgreSQL, containers and Restore and recovery tiers.

When things go wrong: follow Disaster recovery decision rules — farm edge stack for restore order and failure classification.

Canonical package: Backup and disaster recovery package — smart farm stack.

Longhorn operator docs (captures): Longhorn system backup, Longhorn system restore, homelab backup stack source note. k3s / Rancher pointers: backup-dr link batch.

Parallel tracks (do not conflate)

Track Protects Mechanism
A. Application data farmOS DB, broker state Logical dumps (pg_dump class) via restic --stdin-from-command (restic capture); see comparison page
B. Longhorn volumes PVC block data Volume backups + CSI snapshots; system backup bundle for Longhorn metadata (backup system capture)
C. Cluster etcd / k3s state Control plane Embedded etcd: k3s backup and restore, etcd-snapshot; external datastore: follow k3s datastore docs
D. Rancher (if installed) Rancher app state on cluster Rancher backup/restore usage; encryption config must be saved outside the operator’s artifacts
E. Configuration and secrets Helm values, GitOps, secrets Not automatically solved by etcd snapshots for every policy—Sealed Secrets / SOPS / external KMS; govern access to snapshot stores that contain Secret objects (Kubernetes Secrets)

Rule: B without A can still lose application consistency if database files were dirty at snapshot time—keep logical exports for farm records unless you accept crash recovery and have restore-tested that path.

Rule: D is in addition to C when Rancher manages clusters; version-match restore targets per Rancher docs.

Optional HA patterns (storage)

  • Longhorn replicated volumes across nodes : survive single node loss on LAN—not off-site DR alone.
  • Backup target to central NAS or object store : off-site copy still needed for site loss.