Your personal operating system for career transition.

Implement Structured Request Tracing

Build a span-based tracing system for LLM requests -- the core data collection layer that feeds monitoring dashboards and post-hoc debugging.

What you are building

Implement a Tracer class that:

1. Creates a root trace -- with a trace_id, feature name, and start timestamp.
2. Supports named spans -- each span has its own start/end timestamps, latency, and attributes.
3. Works as a context manager -- spans auto-complete on __exit__ with elapsed time, even if the body raises.
4. Exports sorted spans -- export() returns a list of dicts sorted by start_ms.

Usage pattern to support

tracer = Tracer(trace_id="req-123", feature_name="document_qa")

with tracer.span("retrieval", index="docs_v2") as ret_span:
    results = retrieve(query)
    ret_span.set("num_results", len(results))

with tracer.span("llm_call", model="claude-3-5-sonnet-20241022") as llm_span:
    response = call_llm(prompt)
    llm_span.set("prompt_tokens", response.usage.input_tokens)

events = tracer.export()  # list of dicts, one per span

Constraints

• Use only the Python standard library.
• All spans must capture span_id, trace_id, name, start_ms, end_ms, latency_ms.
• Context manager must capture elapsed time even if the body raises an exception.