request-tracing

Instrument HTTP/gRPC endpoints with distributed tracing and RED metrics

📖 中文导读:Instrument HTTP/gRPC endpoints with distributed tracing and RED metrics 以下正文为该技能的原始 SKILL.md 内容(英文),上方安装命令可直接使用。

name: request-tracing description: "Instrument HTTP/gRPC endpoints with distributed tracing and RED metrics" triggers: - "trace HTTP requests" - "instrument endpoints" - "distributed tracing" - "trace context propagation" - "gRPC tracing" priority: 1

Request Tracing

Every request needs: a span, RED metrics, and trace context propagation.

Required Attributes

Attribute Example Note
http.method GET, POST Required
http.route /api/orders/:id Template, not actual path!
http.status_code 200, 500 Required
user.tier premium Recommended (not user_id!)

RED Metrics

Rate:     http.requests.count{method, route, status}
Errors:   http.requests.errors{method, route, error_type}
Duration: http.request.duration{method, route} (histogram p50/p95/p99)

Middleware Flow

Request → Extract trace context (W3C traceparent)
        → Start span (method + route)
        → Start timer
        → Call handler
        → Set status_code
        → Record duration histogram
        → If 4xx/5xx: increment error counter
        → End span
        → Inject trace context into response

Context Propagation

Critical: Always propagate to downstream services:

// Outgoing request
propagator.inject(ctx, headers)

Common Mistakes

  1. Using actual path → /api/orders/12345 creates millions of unique values
  2. Missing propagation → Traces break at service boundaries
  3. Only tracking success → Must track errors with error_type

References

  • references/methodology/red-methodology.md
  • references/platforms/{platform}/http.md