You cannot operate what you cannot see. Track the signals that reflect user experience — latency, errors, traffic, and saturation (the four golden signals) — and define SLOs so you know when the system is actually unhealthy, not just busy.
Key ideas
- Measure user-visible latency (p50/p95/p99), error rate, and saturation — not just CPU and memory.
- Define SLOs (e.g. 99.9% of redirects under 50ms) and alert on error-budget burn rate, not raw spikes.
- Combine logs, metrics, and traces to get from "something is wrong" to "here is exactly why".
Compare
| Golden signal | What it tells you |
|---|---|
| Latency | Slow is the new down |
| Errors | Correctness and availability |
| Traffic | Demand vs capacity |
| Saturation | How full the system is |