# services/monitoring

## Grafana admin secret

The Grafana Helm release expects a pre-existing secret named `grafana-admin`
in the `monitoring` namespace. Create or rotate it with:

```bash
kubectl create secret generic grafana-admin \
  --namespace monitoring \
  --from-literal=admin-user=admin \
  --from-literal=admin-password='REPLACE_ME'
```

Update the password whenever you rotate credentials.

## DCGM exporter image

The NVIDIA GPU metrics DaemonSet expects `registry.bstein.dev/monitoring/dcgm-exporter:4.4.2-4.7.0-ubuntu22.04`, mirrored from `docker.io/nvidia/dcgm-exporter:4.4.2-4.7.0-ubuntu22.04`. Refresh it in Zot when bumping versions:

```bash
skopeo copy \
  --all \
  docker://docker.io/nvidia/dcgm-exporter:4.4.2-4.7.0-ubuntu22.04 \
  docker://registry.bstein.dev/monitoring/dcgm-exporter:4.4.2-4.7.0-ubuntu22.04
```

When finished mirroring from the control-plane, you can remove temporary tooling with `sudo apt-get purge -y skopeo && sudo apt-get autoremove -y` and clear `~/.config/containers/auth.json`.