Prometheus

Prometheus is the cloud-native standard for metric collection. Prometheus 3.0 (2025) features a modern UI, Native Histograms, and direct OpenTelemetry (OTLP) ingestion.

When to Use

• Kubernetes: Standard monitoring stack (Prometheus Operator).
• White-box Monitoring: Measuring internal state (heap usage, request count) via endpoints.
• Alerting: Alertmanager handles de-duplication and routing to Slack/PagerDuty.

Quick Start

# prometheus.yml
global:
  scrape_interval: 15s

scrape_configs:
  - job_name: "node"
    static_configs:
      - targets: ["localhost:9100"]

Core Concepts

Time Series Format

Metrics are identified by name and label pairs. http_requests_total{method="POST", handler="/api"}

PromQL

Powerful query language. rate(http_requests_total[5m])

Pull Model

Prometheus scrapes targets. Apps do not push to Prometheus (usually).

Best Practices (2025)

Do:

• Use High-Cardinality wisely: Native Histograms in v3.0 help, but keep labels bounded.
• Use Service Monitors: In K8s, use the Operator’s ServiceMonitor CRD instead of manual config.
• Use OTLP: Ingest OTel metrics directly if you are transitioning standards.

Don’t:

• Don’t use for logs: It is for metrics only. Use Loki for logs.

References

• Prometheus Documentation