What is distributed tracing and how do you propagate context?

Deep dive

Tracing links work units across boundaries:

• Trace = end-to-end request; span = timed unit of work.
• Propagate context through HTTP headers or message brokers.
• Use consistent sampling strategy across services.
• Correlate traces with logs via trace_id.

Examples

W3C traceparent header:

traceparent: 00-4bf92f3577b34da6a3ce929d0e0e4736-00f067aa0ba902b7-01

Common pitfalls

• Missing propagation in async jobs or queues.
• Sampling at the edge but not downstream (broken traces).
• Creating new trace IDs in each service.

Interview follow-ups

• How do you trace background jobs?
• What is baggage and when would you use it?
• How do you link traces with logs and metrics?