Skip to content

Distributed Tracing — Junior

At junior level, focus on this question:

Why doesn't checking each service's own logs separately let you reconstruct what happened to one specific slow request?

Per-service logs, disconnected

flowchart LR LogA["Service A's logs:\n'handled a request\nat 10:00:01.500'"] LogB["Service B's logs:\n'handled a request\nat 10:00:01.600'"] LogC["Service C's logs:\n'handled a request\nat 10:00:01.900'"] Question["Which of these THREE\nlog entries, across\nTHREE different files,\nbelong to the SAME\noriginal user request?"]

Each service logs its own activity independently, with no shared identifier connecting a specific log line in Service A to the corresponding log line in Service B that resulted from the same original request. With many services and high request volume, timestamps alone are nowhere near enough to reliably reconstruct which log entries across which services belong to one specific request — especially when multiple requests are being processed concurrently across all services at once.

The fix: a shared trace ID that follows the request everywhere

flowchart LR Request["Request arrives,\nassigned trace_id=abc123"] --> ServiceA["Service A logs:\ntrace_id=abc123"] ServiceA --> ServiceB["Service B logs:\ntrace_id=abc123\n(SAME id, passed along)"] ServiceB --> ServiceC["Service C logs:\ntrace_id=abc123\n(SAME id, passed along)"]

A single trace ID is generated when a request first enters the system, and every downstream service call propagates this same ID along (typically via an HTTP header) so every service's logs, spans, and timing data can be tagged with it — letting you query "show me everything that happened for trace_id=abc123" and get a complete, ordered picture of the entire request's journey across every service it touched.

🎓 Takeaway: distributed tracing's foundational idea is deceptively simple — attach one shared identifier to a request at its entry point, and make every service pass that identifier along to every downstream call it makes. Everything else in this topic (spans, sampling, collectors) builds on this one propagated identifier.

Test yourself

  1. Why is matching log entries across services by timestamp alone unreliable at even moderate request volume?
  2. What specific piece of information must every service pass along to its downstream calls for tracing to work at all?
  3. If Service B forgets to propagate the trace ID to Service C, what breaks in the resulting trace?

Continue to middle.md.