Distributed Tracing

Logging tells you **What** happened. Tracing tells you **Where** the time was spent. If a request takes 5 seconds, was it slow because of the Database, a slow Network call to the Identity service, or a locked thread in the API? OpenTelemetry provides the answer.

1. Spans and Traces

• Trace: The complete journey of a request.
• Span: A single unit of work (e.g., a SQL query or an HTTP call).

2. Visualizing with Jaeger

Jaeger provides a Gantt-chart view of your request. You can see exactly how long each internal call took. If a SELECT query took 4.5 seconds of a 5-second request, you know exactly where to optimize without looking at a single line of log text.

4. Interview Mastery

Q: "What is the performance overhead of enabling tracing on every single request?"

Architect Answer: "Tracing can add significant overhead because it generates a lot of telemetry data. For high-traffic systems, we use **Sampling**. Instead of tracing 100% of requests, we might only trace 1% (or only trace requests that take longer than 500ms). This gives us enough data to identify trends and bottlenecks without slowing down the production cluster."