Why it matters
In microservices, traces are the primary tool for understanding request flow. Without them, debugging cross-service latency is guesswork.
Advertisement
The architecture
Trace: root span (initial request) + child spans (each service call).
Context propagation: trace ID + span ID in headers (W3C Trace Context or similar).
Advertisement
How it works end to end
Instrumentation: automatic via libraries (OpenTelemetry auto-instrumentation) or manual span creation.
Sampling: keep all traces expensive; sample 1-10% typical. Tail-based sampling keeps traces with errors or high latency.
Visualization: waterfall shows time each service takes. Bottlenecks visible instantly.