Prometheus 3.11.2

Prometheus scrapes metrics from HTTP endpoints and stores them as time series. It does not collect metrics itself and it does not draw graphs, and almost all early confusion comes from expecting it to do both.

The three pieces

Exporters expose metrics over HTTP. node_exporter covers the host; there are exporters for databases, web servers, and most common services.

Prometheus scrapes those endpoints on a schedule and stores the results.

Grafana queries Prometheus and draws dashboards.

Configuration

# prometheus.yml
global:
  scrape_interval: 15s

scrape_configs:
  - job_name: 'nodes'
    static_configs:
      - targets: ['server1:9100', 'server2:9100']

Set retention explicitly rather than letting it grow:

--storage.tsdb.retention.time=90d

Ninety days of a few hosts is a small number of gigabytes, because the storage format compresses time series aggressively. People consistently overestimate this.

Alerting is the point

groups:
  - name: basics
    rules:
      - alert: DiskAlmostFull
        expr: node_filesystem_avail_bytes{mountpoint="/"} / node_filesystem_size_bytes{mountpoint="/"} < 0.10
        for: 10m
      - alert: HostDown
        expr: up == 0
        for: 5m

Alertmanager handles routing to email, ntfy, or a messaging service.

Alert on symptoms rather than causes: disk nearly full, service down, certificate expiring, backup did not complete. Alerting on every metric that moves produces noise you will start ignoring, which is worse than no alerting.

Running it

Docker Compose with Grafana alongside is the easiest path, and our compose generator produces the file. Our monitoring guide covers the whole stack.

Verify Your Download

Release archives on prometheus.io/download are published with SHA-256 checksums.