Skip to content

Distributed Tracing — Middle

At middle level, focus on this question:

How do trace ID, span ID, and parent-span ID combine to reconstruct the exact call tree of a distributed request?

Prerequisite: junior.md.

Spans: one per unit of work, with parent-child relationships

flowchart TD Span1["Span 1: Service A\ntrace_id=abc, span_id=1,\nparent=none (root)"] Span2["Span 2: Service B\ntrace_id=abc, span_id=2,\nparent=1"] Span3["Span 3: Service C\ntrace_id=abc, span_id=3,\nparent=2"] Span1 --> Span2 --> Span3

A span represents one unit of work (an HTTP handler, a database query, a function call you choose to instrument) with a start time, end time, and metadata. Every span carries the shared trace ID plus its own unique span ID and a parent span ID pointing to whichever span caused it — this parent-child chain is what lets a tracing backend reconstruct the exact call tree (not just a flat list of events) for the whole request.

Context propagation over HTTP

GET /api/orders/42 HTTP/1.1
traceparent: 00-abc123def456-789xyz-01

The W3C Trace Context standard defines the traceparent header format (version-traceID-parentSpanID-flags) — when Service A calls Service B, it includes this header; Service B's tracing library extracts it, creates a new span with parent = <the incoming parentSpanID>, and propagates its own new span ID onward to any calls it makes, continuing the chain.

sequenceDiagram participant ServiceA participant ServiceB ServiceA->>ServiceA: start span_id=1 (root) ServiceA->>ServiceB: HTTP call, traceparent header\nincludes trace_id + span_id=1 ServiceB->>ServiceB: start span_id=2, parent=1 Note over ServiceB: If B calls C, B propagates\ntraceparent with parent=2

🎓 Takeaway: the trace ID identifies "which overall request," the span ID identifies "which specific piece of work," and the parent-span ID reconstructs "which piece of work caused which other piece of work" — together, these three pieces of propagated context are sufficient to rebuild the complete, ordered, hierarchical timeline of a request across any number of services.

Test yourself

  1. Why is a parent-span ID necessary, in addition to the shared trace ID — what would be missing if every span only carried the trace ID?
  2. Trace through the traceparent header propagation for a 4-service call chain (A → B → C → D), and identify each span's parent.
  3. Why does the W3C Trace Context standard matter for interoperability between services written in different languages or using different tracing libraries?

Continue to senior.md.