How do I get logs, metrics, and tracing for every API request?

To get logs, metrics, and tracing for every API request, you want observability middleware in your API layer, plus an export pipeline to a backend that stores/visualizes the data.

1) What to capture on every request

For each request, collect:

• Logs: method, path, status code, latency, request ID, user/tenant ID if available, error details
• Metrics: request count, latency histogram, error rate, throughput, payload size
• Traces: a span per request, plus child spans for downstream calls (DB, cache, HTTP, queue)

2) Add request middleware/interceptors

At the API boundary, add middleware that:

1. Starts a trace span
2. Injects/extracts trace context from headers
3. Generates or forwards a correlation/request ID
4. Measures duration
5. Logs structured request/response info
6. Emits metrics to a metrics system

Typical headers:

• traceparent / tracestate for distributed tracing (W3C Trace Context)
• X-Request-Id or similar for correlation

3) Use OpenTelemetry as the standard

The easiest modern approach is OpenTelemetry (OTel):

• Tracing: automatic + manual spans
• Metrics: request counters, latency histograms
• Logs: structured logs correlated with trace/span IDs

Recommended architecture:

• Your app → OpenTelemetry SDK
• OTel SDK → OTel Collector
• Collector → backends:
  • Traces: Jaeger / Tempo / Honeycomb / Datadog / New Relic
  • Metrics: Prometheus / Grafana Mimir / Datadog
  • Logs: Loki / Elasticsearch / Cloud logging

4) Make logs trace-aware

Include these fields in every log line:

• trace_id
• span_id
• request_id
• service
• environment
• route
• status_code
• duration_ms

This lets you jump from a log entry to the exact trace.

5) Instrument downstream dependencies too

To get full tracing, instrument:

• database queries
• cache calls
• outbound HTTP/gRPC requests
• message queue publish/consume

Otherwise you’ll only see the top-level request span.

6) Example pattern

For each incoming request:

• start span: HTTP GET /users/:id
• record attributes:
  • http.method
  • http.route
  • http.status_code
  • enduser.id (if appropriate)
• increment metric: http_requests_total{route,method,status}
• observe latency histogram
• log structured event with trace ID

7) Practical stack examples

Simple stack

• App instrumentation: OpenTelemetry
• Traces: Jaeger or Grafana Tempo
• Metrics: Prometheus + Grafana
• Logs: Loki + Grafana

Managed stack

• Datadog / New Relic / Honeycomb / Azure Monitor / AWS X-Ray + CloudWatch

8) Important best practices

• Don’t log secrets, tokens, passwords, or full PII
• Sample traces if traffic is high, but keep error traces always
• Use structured JSON logs
• Propagate context across async jobs and background workers
• Add a unique request ID at the edge (gateway/load balancer)

9) If you want “every request” specifically

You need all three of these:

• Always log a request summary
• Always count request metrics
• Always start a trace span for incoming requests

If you tell me your stack (e.g. Node/Express, Python/FastAPI, Go, Java/Spring, .NET, Kubernetes, AWS, etc.), I can give you a concrete setup and code example.