Monitoring and Observability
Monitoring and Observability
You Cannot Fix What You Cannot See
Once applications run across many containers, something will go wrong: a pod restarts, a database slows, or latency spikes. Monitoring and observability are how you notice, understand, and react. Monitoring asks whether something is wrong, and observability asks why it is wrong, using rich data collected ahead of time.
The Three Pillars
Observability rests on three well-known signals:
- Metrics: numeric measurements over time, such as CPU usage, request rate, and error rate.
- Logs: timestamped text records of events, useful for chasing the details of a single request.
- Traces: records of a request's path through services, showing which component caused a slow response.
Metrics with Prometheus
Prometheus is the standard metrics collector. Applications expose a metrics endpoint, Prometheus scrapes it on schedule, and a dashboard such as Grafana visualizes the numbers.
# HELP http_requests_total Total HTTP requests
# TYPE http_requests_total counter
http_requests_total{method="GET"} 1024
That sample text format is easy for applications to produce, which is why so many libraries ship Prometheus endpoints by default.
Dashboards and Alerts
Dashboards turn raw numbers into graphs, but graphs only help when someone looks at them. Alerting closes the loop: rules watch metrics and page a human when a threshold is breached, such as an error rate above a few percent for five minutes. Every alert should say what is wrong and where to look, and it should be worth waking someone up.
Health Checks
Load balancers and orchestrators use health checks to decide if a service is fit to receive traffic. A readiness probe tells Kubernetes a container is ready; a liveness probe tells it the container is alive and should be restarted if it is not. Wiring both into your deployments is the cheapest, highest-value monitoring you can add.
Key Points
- Monitoring detects problems; observability explains them.
- Metrics, logs, and traces are the three observability pillars.
- Prometheus collects metrics; Grafana visualizes them.
- Alerts must be actionable and only page for real incidents.
- Readiness and liveness probes let orchestrators supervise containers.
