Conclusion

​

Course summary

​

Practical next steps

1. Instrument a small service (HTTP server, microservice, or CLI) using a Prometheus client library.
2. Deploy a Prometheus instance and add your service as a scrape target.
3. Create basic PromQL queries and visualize them in Grafana or the Prometheus UI.
4. Add alerting rules and integrate Alertmanager for notification routing.
5. Iterate on label design and monitor cardinality and storage costs.

# Rate of HTTP requests per second across all instances
sum by (job) (rate(http_requests_total[5m]))

# 95th percentile request duration for a particular service
histogram_quantile(0.95, sum(rate(http_request_duration_seconds_bucket[5m])) by (le, service))

​

Best practices & considerations

• Design labels carefully: prefer stable, low-cardinality labels such as job, service, and region. Avoid high-cardinality labels like unique request IDs.
• Keep retention and storage trade-offs in mind: longer retention increases cost and storage requirements.
• Use exporters for third-party systems (node_exporter, kube-state-metrics, Blackbox exporter) rather than custom scraping where possible.
• Protect your Prometheus server: secure access to the UI, consider remote write to long-term storage, and set resource limits for scaling.

​

Further reading & references

• Prometheus Documentation
• PromQL Guide
• Alertmanager
• Grafana
• Prometheus Exporters