TopolearnDashboards are what most people use Grafana for, and a few good habits make them readable from day one.
Create a dashboard with Dashboards, New, New dashboard, Add visualization, and pick the data source. The panel editor has three areas: the preview, the query section below it, and the visualization options on the right.
For Prometheus the query editor has a Builder mode (pick metric, labels and operations) and a Code mode for typing PromQL. Useful query options:
- Legend:
{{instance}}or{{status}}to name series by label instead of the full label set. - Min step / interval and Type (range, instant, both). Stat and table panels often need instant queries.
Common visualizations: Time series for trends, Stat for one big number with a sparkline, Gauge and Bar gauge for values against limits, Table for lists, Heatmap for histogram distributions.
The dashboard's time picker (top right) sets the range for every panel, such as "Last 6 hours", plus an auto-refresh interval. Drag panels to arrange them and group them in collapsible rows.
Save the dashboard into a folder with a meaningful name. Each save creates a version you can compare and restore under Settings, Versions. Start with a few signals that answer "is the service healthy?" (traffic, errors, latency) rather than every metric you have.
Example
# Panel 1: Time series "Requests per second", legend {{status}}
sum by (status) (rate(http_requests_total{job="api"}[$__rate_interval]))
# Panel 2: Stat "Error ratio", unit Percent (0.0-1.0), instant query
sum(rate(http_requests_total{job="api", status=~"5.."}[5m]))
/ sum(rate(http_requests_total{job="api"}[5m]))
# Panel 3: Time series "p95 latency", unit seconds
histogram_quantile(0.95,
sum by (le) (rate(http_request_duration_seconds_bucket{job="api"}[$__rate_interval])))
# Panel 4: Stat "Targets up", instant query
sum(up{job="api"})Remember
- Panels = query + visualization + options
- Use legend templates like {{status}} and instant queries for stats
- Save into folders; every save is a restorable version
Watch out for
- One dashboard with 60 panels of every metric; nobody can read it during an incident.