Distributed Tracing (Jaeger, OpenTelemetry)
In microservices and distributed systems, a single user request may touch dozens of services.
Distributed tracing provides end-to-end visibility into these requests.
1. Why Distributed Tracing?
- Debug latency across multiple services.
- Find bottlenecks in service calls.
- Correlate logs/metrics with request flow.
- Improve MTTR (mean time to recovery).
2. How Tracing Works
- Each request assigned a trace ID.
- Each service call within request → span.
- Trace = collection of spans showing the full request path.
Example
Trace ID: 12345
Span A: API Gateway (50ms)
Span B: Auth Service (20ms)
Span C: DB Query (200ms)3. Tools
Jaeger
- Open-source, CNCF project.
- Collects, stores, and visualizes traces.
- Integrates with Kubernetes, Istio.
OpenTelemetry
- Open standard for telemetry data.
- Supports traces, metrics, logs.
- Vendor-neutral → export to Jaeger, Datadog, etc.
Zipkin
- Early open-source tracing tool.
- Still used, but less popular vs Jaeger.
4. Real-World Usage
- Uber → built Jaeger for large-scale tracing.
- Google Dapper → research paper that inspired modern tracing.
- Istio Service Mesh → built-in tracing with OpenTelemetry.
5. Best Practices
- Propagate trace IDs across all services.
- Use correlation IDs in logs.
- Sample traces (don’t trace every request in high-QPS systems).
- Combine with metrics + logging for full observability.
6. Interview Tips
- Say: “I’d use distributed tracing (Jaeger, OpenTelemetry) to debug latency across services.”
- Mention trace ID + spans explicitly.
- Show awareness of performance overhead.
7. Diagram
[ Client Request ] → [ API Gateway (Span 1) ]
↓
[ Service A (Span 2) ] → [ Service B (Span 3) ]
↓
[ Database (Span 4) ]8. Next Steps
- Learn about Alerting Systems.
- Explore Soft Skills for HLD Interviews.