Skip to content

Grafana Fundamentals: Data Sources, Panels & Dashboards

What was run

A real local Grafana 13.2.4 instance (port 3000), connected to the real local Prometheus server from Level 1 as a real data source, with a real dashboard and panel created through Grafana's own HTTP API and verified by actually loading it in a browser and watching it render real data.

A real data source, actually tested

curl -s -X POST http://admin:admin@localhost:3000/api/datasources -H "Content-Type: application/json" -d '{
  "name": "Prometheus", "type": "prometheus", "url": "http://localhost:9090", "access": "proxy", "isDefault": true
}'

Adding the data source through the API alone doesn't prove it works — clicking "Save & test" on the real data source settings page in a real browser is what actually exercises the connection, and it returned a real, direct confirmation:

✓ Successfully queried the Prometheus API.

A real panel, rendering real data

curl -s -X POST "http://admin:admin@localhost:3000/api/dashboards/db" -H "Content-Type: application/json" -d '{
  "dashboard": {
    "title": "Orders Service Overview",
    "panels": [{
      "type": "timeseries", "title": "Checkout request rate",
      "gridPos": {"h": 8, "w": 12, "x": 0, "y": 0},
      "datasource": {"type": "prometheus", "uid": "ag2egat7m3l6oe"},
      "targets": [{"expr": "sum(rate(orders_requests_total{endpoint=\"/checkout\"}[1m]))", "refId": "A"}]
    }]
  }
}'

Loading this dashboard in a real browser rendered a real line graph with two visible spikes, sitting at 0 between them and rising to ~1 during each — exactly matching the two real bursts of 50 checkout requests sent to orders_app.py across Level 1, Lesson 10. The panel's legend showed the exact real PromQL expression it was built from, and the shape of the graph is a direct, visible encoding of real, observed traffic — not a static mockup.

How It Actually Works

A Grafana data source is a configured connection to wherever your real data lives — Prometheus here, but also logs backends, other time-series databases, or SQL databases. A panel is one visualization (a graph, a single number, a table, a heatmap) built from one or more real queries against a data source, and a dashboard is a collection of panels, laid out together, usually because they answer related questions about the same real system.

The real, important distinction from a static screenshot or a hand-drawn mockup: every panel on a real Grafana dashboard re-runs its real query against the real data source on a schedule (or whenever the dashboard's time range changes), which is exactly what made the two real spikes above show up automatically — nobody manually plotted those points; Grafana queried Prometheus, got back real data, and rendered it. This is also why a dashboard's correctness depends entirely on its underlying PromQL being correct — a dashboard is only as trustworthy as the real query driving each panel, which is why Level 1's PromQL foundations matter before dashboards can be built reliably on top of them.

Grafana's dashboards are themselves stored as JSON (the panels array shown in the API call above), which is what makes Level 4, Lesson 7's "dashboards as code" possible later in this course — a dashboard built once can be exported, version-controlled, and recreated exactly through the same API this lesson used directly.

Common mistakes

  • Trusting that a data source is configured correctly just because adding it didn't show an error, without actually clicking "Save & test" (or querying through it) to get a real, direct confirmation the connection works end to end.
  • Building a panel with a query that's syntactically valid PromQL but doesn't actually answer the question the panel's title claims to — Grafana will happily render whatever a query returns, correct or not.
  • Treating a dashboard screenshot (shared in a chat message, say) as equivalent to the live dashboard — a screenshot is frozen at the moment it was taken, while the real dashboard keeps re-querying and updating.

Exercise

Add a second panel to this dashboard, a real stat panel (a single big number, not a graph) showing the current real value of orders_inflight_requests from Level 1, Lesson 3. Load the dashboard in a real browser, fire a real burst of concurrent requests at orders_app.py the way that lesson did, and confirm you can watch the real number change live as the burst happens and completes.