Node with Metrics

A minimal Node.js + Express service demonstrating production-ready Prometheus metrics alongside OpenTelemetry distributed tracing, with exemplars bridging the two.

Quick start

nvm install $(cat .nvmrc)
corepack enable
yarn install
yarn dev

What's instrumented

Prometheus metrics (via prom-client, hand-crafted):

• http_request_duration_seconds — histogram with trace ID exemplars
• http_requests_total — counter by method, route, status code
• http_requests_in_flight — gauge
• http_response_size_bytes — histogram
• items_processed_total — business metric with status labels
• app_info / app_start_timestamp_seconds — metadata gauges
• Default Node.js runtime metrics (GC, event loop, memory, etc.)

OpenTelemetry traces (auto-instrumented):

• Express route handling
• HTTP client/server spans
• Trace context propagation across services

Exemplars link the two: each histogram observation carries the OTel trace ID, so you can click from a Prometheus latency spike directly into the Jaeger trace that caused it.

Grafana dashboards:

• HTTP Metrics
• Node.js Runtime

Full observability stack

docker compose up

Service	URL
App	http://localhost:3000
Prometheus	http://localhost:9090
Jaeger	http://localhost:16686
Grafana	http://localhost:3001

Grafana is setup to provision the Prometheus and Jaeger datasources. Exemplar linking is configured — query a histogram in Grafana and click the trace ID to jump to Jaeger.

Endpoints

Method	Path	Purpose
GET	/	Health check
GET	/version	Package version
GET	/metrics	OpenMetrics/Prometheus
GET	/healthz	Kubernetes liveness probe
GET	/readyz	Kubernetes readiness probe
POST	/items	Simulated async work
GET	/error	Always returns 500 (for dashboard demos)

/metrics, /healthz, and /readyz are excluded from metrics recording.

Development

yarn install      # Install dependencies
yarn dev          # Dev server with tracing + hot reload
yarn build        # Compile TypeScript
yarn test         # Run tests
yarn lint         # ESLint + Prettier