Raspberry Pi k3s fleet — backup and restore sequence¶
Parent runbook: How to provision k3s, Longhorn, and Rancher on a Raspberry Pi fleet. DR package hub: Backup and disaster recovery package — smart farm stack. DR decision rules: Disaster recovery decision rules — farm edge stack. Platform layer: Kubernetes platform backup / DR — Pi, k3s, Longhorn. Granularity compare: Backup strategy comparison — farmOS, homelab, PostgreSQL, containers.
Three tracks (understand all; implement per phase)¶
| Track | Protects | P0 | P1 | Later |
|---|---|---|---|---|
| A. Application logical dumps | Coherent DB exports (e.g. pg_dump) |
Optional | Mandatory for farm records | Mandatory |
| B. Longhorn volume / system backup | PVC block data + Longhorn metadata | Optional | Recommended | Mandatory with tested restore |
| C. etcd / k3s state | Control plane | Rarely needed if cluster is throwaway | Document method | Mandatory with HA story |
Rule (from the platform page): track B without track A can still lose application consistency if the database was not quiesced or logically exported.
Mandatory — P1 operator sequence (outline)¶
- Choose a backup destination reachable from the cluster (S3-compatible API, NFS on a NAS, restic to external disk, etc.).
- Implement track A for each stateful application (cron + restic
--stdin-from-commandpattern per the comparison page). - Configure Longhorn backup targets and policies per Longhorn docs (see vault captures in source notes) after base volumes work.
- Document restore order in a tabletop exercise before you need it: PVC restore, namespace recreate, DB import order.
- Record etcd / k3s snapshot approach matching your install mode (single server vs HA) using k3s datastore documentation.
Optional (HA / scale)¶
- Off-site replication of restic repos or versioned object storage.
- Longhorn system backup bundle for storage-layer DR (not a full cluster clone by itself).
Restore drill flow (reference)¶
mermaid
flowchart TD
L[Detect loss] --> T{Which track failed?}
T -->|App| A[Restore DB from dump]
T -->|PVC| B[Longhorn restore per docs]
T -->|Cluster| C[etcd / k3s recovery]
A --> V[Validate app]
B --> V
C --> V
Pi / off-grid notes¶
- Schedule heavy backup jobs when power margin exists —
Off-grid implications — backup and networking choices. - When using restic with database dumps, follow upstream guidance so dump failures are not masked (see
Backup strategy comparison).