Arqly
Sign in
#31Build a distributed tracing system (Jaeger / Zipkin style)
stub
Logs tell you what happened in one service; metrics tell you the rate; only traces tell you the full causal chain across N services for one user request. Build the third pillar of observability: span propagation, sampling that doesn't lie, span ingestion, trace storage by trace-id, and the UI flame graph that finally answers 'where did the latency go?'.

Build a distributed tracing system (Jaeger / Zipkin style). Logs tell you what happened in one service; metrics tell you the rate; only traces tell you the full causal chain across N services for one user request. Build the third pillar of observability: span propagation, sampling that doesn't lie, span ingestion, trace storage by trace-id, and the UI flame graph that finally answers 'where did the latency go?'.

This problem is a stub. Suggested approaches haven't been authored yet — ask the staff engineer in the right panel for any stage and they'll generate one tuned to your draft.
Reading: Sigelman et al. — Dapper, a Large-Scale Distributed Systems Tracing Infrastructure (Google 2010) · Cindy Sridharan — Distributed Tracing in Practice (O'Reilly) · OpenTelemetry specification — Tracing API, propagators, SDK · Jaeger documentation — Architecture, sampling, storage backends · Zipkin documentation — Architecture, instrumented frameworks · Honeycomb engineering blog — Why columnar storage for traces · Lightstep / ServiceNow — Tail-based sampling design papers
spans, traces, and the parent-child relationship
context propagation: W3C Trace Context, B3 headers
sampling: head-based vs tail-based (and why head is what everyone ships)
sampling biases: random, rate-limit, probability, adaptive
OpenTelemetry as the polyglot collection standard
span batching, compression, and OTLP export
ingest pipeline: collector → buffer → processor → storage
trace storage: index by trace-id (point lookup), wide rows, retention
trace-aware flame graphs + critical path analysis
tail-based sampling: keep traces with errors / high latency
the cost of full-fidelity tracing (10–30% of compute on hot paths)
No scenes authored for this problem yet.