Skip to content

Grafana Dashboards

The Gateway repository maintains monitoring assets under deploy/monitoring. Use assets from the same selected source as the metric contract. A dashboard JSON file establishes its queries and layout; it does not prove that a deployed panel receives data or that an alert threshold fits your workload.

Verify the data path

  1. Enable the required metric producers using Monitoring and Metrics.
  2. Check a direct Gateway scrape, then the Prometheus target and scrape errors.
  3. Check Grafana's selected data source, instance filter, time range, and query response. An empty panel is not a zero measurement.
  4. Generate owned representative traffic and compare counter deltas with independently recorded request outcomes.

The generated metric reference records exact family names, label schemas, producer activation, and upstream evidence classes. Interpret each measurement within that scope. Do not combine host CPU and container CPU, aggregate and model quota utilization, or request count and transport connection count as though they were the same measurement.

Access and changes

Restrict dashboard and data-source access to the intended operators. Review labels and exported evidence for sensitive tenant, model, and endpoint data. Keep a copy of reviewed dashboard changes outside disposable containers. Validate notification delivery and workload-specific thresholds independently before relying on an alert for operations.