Core Concepts — The Car Dashboard Blueprint

Before the mechanic can install the dashboard, everyone needs to learn the vocabulary. Grafana has a small set of core concepts — once you understand these, every dashboard configuration makes sense.

The Big Picture

Data Sources — Where the Data Comes From

A Data Source is a connection to a monitoring system. Grafana doesn't collect data — it reads it from external sources.

Data Source	What it provides
Prometheus	Metrics and time series
Loki	Logs
Tempo	Traces
Elasticsearch	Search and log analytics
InfluxDB	Time series metrics
MySQL/PostgreSQL	Relational data

Grafana provisioning — add Prometheus as data source

apiVersion: 1

datasources:
  - name: Prometheus
    type: prometheus
    url: http://prometheus:9090
    access: proxy
    isDefault: true

Remember

Grafana is a visualization layer. It reads from data sources but doesn't store metrics. Think of it as the glass on the car dashboard — it displays data from the engine sensors.

Dashboards — The Instrument Cluster

A Dashboard is a collection of panels arranged in a grid. It's the full instrument cluster — speedometer, fuel gauge, temperature gauge, all in one place.

Each dashboard has:

• Panels — individual visualizations (graphs, gauges, tables)
• Rows — logical groupings of panels
• Variables — dropdowns that filter all panels
• Time range — the period being displayed

Remember

A dashboard is not a query. It's a collection of panels, each with its own query. The dashboard time range applies to all panels unless overridden.

Panels — The Individual Gauges

A Panel is a single visualization — a graph, gauge, stat, table, or text display. Each panel has its own PromQL query (or query from any data source).

Panel Type	Best for	Example
Time series	Trends over time	CPU usage over 1 hour
Stat	Single current value	Current error count
Gauge	Value within a range	Memory usage (0-100%)
Table	Structured data	Top 10 slowest endpoints
Bar gauge	Comparing values	Requests per service
Heatmap	Distribution over time	Request latency distribution
Alert list	Active alerts	Current firing alerts

Panel anatomy

Each panel consists of:

1. Query — the PromQL (or SQL, etc.) expression
2. Visualization — how to display the data
3. Transformations — modify the data before display
4. Alert — optional threshold-based alerting

Remember

One panel = one visualization. If you need to compare two metrics, use two panels or use a legend to show both lines in one graph.

Variables — The Dashboard Controls

Variables are dropdown menus that let you filter all panels on a dashboard. They're the controls on the car dashboard — select which lane, which engine, which time range.

Template variable — select a server

templating:
  list:
    - name: instance
      type: query
      query: label_values(up, instance)
      datasource: Prometheus

Now $instance can be used in any panel query:

Use the variable in a query

rate(http_requests_total{instance="$instance"}[5m])

Remember

Variables make dashboards reusable. One dashboard with an instance variable works for every server.

Rows and Layout — Organizing the Dashboard

Rows group related panels visually. Click a row to collapse/expand it.

Dashboard layout

panels:
  - title: Overview
    type: row
    collapsed: false

  - title: CPU Usage
    type: timeseries
    gridPos: { h: 8, w: 12, x: 0, y: 1 }    # half width, row 1

  - title: Memory Usage
    type: timeseries
    gridPos: { h: 8, w: 12, x: 12, y: 1 }   # half width, right side

  - title: Request Rate
    type: timeseries
    gridPos: { h: 8, w: 24, x: 0, y: 9 }    # full width, row 2

Remember

gridPos defines the panel position: h = height, w = width (out of 24), x = horizontal position, y = vertical position.

Annotations — Markers on the Timeline

Annotations are vertical markers on graphs that highlight specific events (deploys, incidents, maintenance).

Annotation from Prometheus

annotations:
  - name: Deployments
    datasource: Prometheus
    query: 'count_over_time(deployments_total[1m]) > 0'

Remember

Annotations correlate events with metrics. "The spike happened right after the deploy" becomes visible instantly.