Application Performance Monitoring (APM) Setup Checklist

Application is slow or unreliable in production but there's no visibility into where time is spent, which requests are failing, or what's degrading.

Essential APM instrumentation:

1. Request tracing (most important)

• Trace every HTTP request: method, path, status, duration
• Add trace IDs that propagate across services
• Sample high-traffic endpoints (e.g., 10% of GET /api/feed)

2. Key metrics (RED method)

• Rate: requests per second
• Errors: error rate percentage
• Duration: p50, p95, p99 latency

3. Database monitoring

• Query duration and count
• Slow query log (>100ms)
• Connection pool utilization
• N+1 query detection

4. External dependency tracking

• API call duration and error rates
• Circuit breaker state
• Timeout frequency

5. Custom business metrics

• Sign-ups per minute
• Orders processed
• Payment success/failure rate
• Queue depth and processing time

6. Alerting thresholds

# Alert on symptoms, not causes
alerts:
  - name: High Error Rate
    condition: error_rate > 1% for 5m
    severity: critical
  - name: High Latency
    condition: p99_latency > 2s for 5m
    severity: warning
  - name: Saturation
    condition: cpu_usage > 80% for 10m
    severity: warning

Tools: Datadog, New Relic, Grafana + Prometheus, Sentry (errors), Jaeger (tracing).

You can't fix what you can't see. APM transforms debugging from 'something is slow' to 'the /api/checkout endpoint p99 is 3s because PostgreSQL query on line 42 takes 2.8s'.

Setting up production monitoring