# Data Sources and Panels — Prometheus + Grafana

Source: https://www.skillbyai.com/en/prometheus-grafana/g-panels

> From query to visualisation.

## Connect Prometheus, then build panels

Add Prometheus as a **data source** (its URL, for example http://prometheus:9090), then create a dashboard of **panels**. Each panel runs one or more queries and shows them as a time series graph, stat, gauge, bar chart, table or heatmap (good for histogram buckets). Set the correct **unit** (seconds, bytes, percent, requests per second) and thresholds so colours highlight problems. **Explore** mode is for ad-hoc querying during investigations, and Grafana can also alert on queries.

## Visualise and explore

Grafana turns PromQL queries into dashboards that help people understand systems quickly.

![Three ideas: data sources and panels, variables and good layout, dashboards as code.](assets/figures/prometheus-grafana/section-7-map.svg) — Figure 7.1 — Panels, variables and provisioning.

## Panels for a service overview

Queries for a RED dashboard.

```text
Panel "Request rate"   (time series, unit req/s)
  sum by (route) (rate(http_requests_total{job="$job"}[$__rate_interval]))
Panel "Error ratio"    (stat, unit percent 0-1, thresholds 1% amber, 5% red)
  job:http_requests_error_ratio:rate5m{job="$job"}
Panel "p95 latency"    (time series, unit seconds)
  histogram_quantile(0.95, sum by (le) (rate(http_request_duration_seconds_bucket{job="$job"}[$__rate_interval])))
Panel "Latency heatmap" (heatmap, format: heatmap)
  sum by (le) (increase(http_request_duration_seconds_bucket{job="$job"}[$__rate_interval]))
```

## Use $__rate_interval

Grafana's $__rate_interval adapts the rate window to the zoom level and scrape interval, avoiding empty or jagged graphs.

**Quiz:** Which panel type best shows a latency distribution over time?

- [ ] Text
- [ ] Stat
- [ ] Pie chart
- [x] Heatmap

*Answer:* Heatmap. Buckets over time.