Micrometer와 Prometheus로 메트릭 및 대시보드 구축

Logs tell you what happened and traces show request paths, but metrics reveal trends over time: request rates, error percentages, memory use. They complete the observability picture.

Micrometer is a metrics facade, like SLF4J but for metrics. You instrument once and export to many backends such as Prometheus or Datadog.

Add the Prometheus registry dependency and Actuator exposes a scrape endpoint automatically.

Enable the Prometheus endpoint so metrics are served at /actuator/prometheus in a format Prometheus understands.

Most metrics fall into three types.

• Counter — only goes up (requests served)
• Gauge — goes up and down (queue size)
• Timer — measures durations

Register a counter and increment it when an event occurs.

A Timer records how long operations take, giving you latency percentiles.

Add tags (labels) to slice metrics by endpoint, status, or region. Avoid high-cardinality tags like user IDs, which explode storage.

Prometheus periodically pulls the /actuator/prometheus endpoint and stores time-series data, which you can then query with PromQL.

Grafana connects to Prometheus as a data source and renders dashboards. Spring Boot has ready-made dashboards for JVM and HTTP metrics.

Define alert rules, for example error rate above 5 percent, so the team is notified before users complain. Metrics turn into proactive operations.

You exposed metrics via Micrometer and the Prometheus Actuator endpoint, learned counters, gauges, and timers, used tags wisely, and visualized and alerted with Prometheus and Grafana.