Deploy in-cluster

This is the recipe for an in-cluster Grafana (e.g. kube-prometheus-stack). The repo's deploy/ folder has everything.

1. Read-only ServiceAccount + RBAC

kubectl apply -f deploy/rbac.yaml

Creates a kubemeridian-readonly ServiceAccount with a ClusterRole granting get/list/watch on the resources the app reads (nodes, pods, namespaces, services, workloads, events, PVC/PV, ingresses, resourcequotas, HPAs) — and a long-lived token Secret.

2. Install the plugin (Helm overlay)

helm upgrade <kps-release> prometheus-community/kube-prometheus-stack \
  -n monitoring --reuse-values -f deploy/kps-grafana-overlay.yaml

The overlay only touches grafana.*:

• GF_INSTALL_PLUGINS — installs the unsigned zip: the v2.5.2 release asset, or a zip you build and host yourself.
• allow_loading_unsigned_plugins — both ids (devopstech-kubemeridian-app, devopstech-kubemeridian-datasource); in Grafana 13 the bundled datasource does not inherit the app's permission.
• deploymentStrategy: Recreate — the Grafana PVC is RWO; RollingUpdate deadlocks on Multi-Attach.

3. Provision the datasource

The datasources sidecar watches Secrets labelled grafana_datasource=1. Inject the SA token (not committed) and apply:

TOKEN=$(kubectl -n monitoring get secret kubemeridian-readonly-token \
          -o jsonpath='{.data.token}' | base64 -d)
sed "s|__SA_TOKEN__|$TOKEN|" deploy/grafana-datasource.yaml | kubectl apply -f -

Provisioned datasources are API-read-only — edit connections by changing this Secret and re-applying, not via the UI.

4. Enable the app & install alerts

curl -s -u admin:PASS -X POST -H 'Content-Type: application/json' \
  -d '{"enabled":true,"pinned":true}' \
  https://<grafana>/api/plugins/devopstech-kubemeridian-app/settings

kubectl apply -f deploy/kubemeridian-prometheusrule.yaml   # optional alert pack

Replacing the installed zip

GF_INSTALL_PLUGINS skips an already-installed plugin, so to switch to a different zip (for example one you built yourself): point the overlay at it, then rm -rf /var/lib/grafana/plugins/devopstech-kubemeridian-app in the pod and rollout restart the Grafana deployment.