Grafana¶

Purpose¶

kubernetes/apps/observability/grafana runs grafana-operator plus a single Grafana instance CR. Dashboards ship as declarative GrafanaDashboard custom resources across the repository, and the Prometheus and Alertmanager datasources are wired in directly rather than through the UI.

Backups¶

The Grafana CR's persistentVolumeClaim block creates and owns the data volume itself (grafana-pvc), rather than through the app-template persistence pattern most other apps use.

instance/kustomization.yaml references components/volsync/nfs and components/volsync/remote directly instead of the parent components/volsync component. The parent component also ships pvc.yaml, an app-template-style claim named ${APP} (grafana). Pulling that in here would create a second, unused PVC that would try to restore from a ReplicationDestination with no snapshot yet, since nothing ever backs up to it.

Both ReplicationSource templates default sourcePVC to ${APP}, which resolves to grafana, not the operator's real claim. instance/kustomization.yaml patches that field locally to grafana-pvc rather than changing the shared component.

What actually needs backing up is state that Git does not hold: the operator's SQLite database, local users, preferences, annotations, and alert-rule state, plus any dashboard built ad hoc in the UI instead of committed as a GrafanaDashboard CR. The dashboards already tracked in this repository are declarative and do not depend on a PVC snapshot to recover.