Skip to content

Monitoring and logging expectations — edge cluster standard

Purpose

Minimum observability bar for farm edge k3s clusters: enough signal to detect disk, API, and workload failure without mandating a specific vendor stack.

Not in scope: Full Prometheus architecture—pick components per Platform decision memo.

Standard

Area Expectation
Cluster health Someone can answer: API up? Nodes Ready? Longhorn volumes attached?—via kubectl, dashboard, or metrics.
Stateful workloads farmOS / DB: HTTP or app probe outside “pod running”—verify read/write path periodically.
Logs Retain policy documented (days/weeks); centralize if multi-node—avoid only ephemeral container logs for debugging tax events.
Alerts At least one channel (email, phone, HA notify) for backup job failure and cert expiry if ingress used.
Pi / power On off-grid sites, pair with SOC / generator awareness (Off-grid implications — backup and networking choices)—not only K8s metrics.

Pilot vs production: Pilot may be manual kubectl only; production implies scheduled checks or metrics scraping.