Metrics

Metrics summarize behavior over time as counters, gauges, histograms, and rates. They power SLIs, capacity planning, alerting, and regression detection after deploys.

| RED (services) | USE (resources) | |----------------|-----------------| | Rate | Utilization | | Errors | Saturation | | Duration | Errors |

					httpRequestDuration.observe(
  { method: 'GET', route: '/api/orders', status: '200' },
  elapsedSeconds,
);
				

Control label cardinality — do not put unbounded user IDs on metric labels.

On interviews: latency percentiles, error rate, saturation, RED/USE, cardinality limits, and why averages hide tail latency.

Common pitfalls: high-cardinality labels overwhelm systems; dashboards without SLOs become unused wallpaper.

The trade-off is metric granularity versus storage cost and alert noise.

Checklist:

• Track latency, errors, traffic, saturation.
• Control label cardinality.
• Tie metrics to SLOs and ownership.
• Compare deploy windows for regressions.