SlackTrace: A New Tracing Tool
summary
SlackTrace introduces a minimal SpanEvent span format and a pipeline (Neuron → Kafka → adapters → Honeycomb + Presto) that stores full-fidelity spans and lets teams run SQL over raw span rows. It traces 100% of dev requests and about 1–2% sampling in prod, supports Zipkin/Jaeger adapters, and powers ad‑hoc joins that found memcache p99 regressions, proved or disproved cross‑region IDR latency, modeled WebSocket state, and proved effective in production triage.
AI-written summary · about
chapters (18)
- 0:00Intro
- 0:28About Suman
- 2:23Tracing Systems
- 3:22Complex Systems
- 5:55The Problem
- 7:17Why Do We Care
- 8:23Tracing Today
- 12:29Query Languages
- 13:28New Events
- 14:35Spans
- 15:53Feet Tracing
- 16:07Pipeline Infrastructure
- 17:37Request View
- 17:59Heat Maps
- 18:09Sample Query
- 25:00Future plans
- 26:29Summary
- 27:14Questions
related talks
23mHow Tracing Uncovers Half-truths in Slack’s CI Infrastructure
49mDistributed Tracing: How the Pros Debug Concurrent and Distributed Systems
54mLatency and Event Tracing with Spring Cloud Data Flow
45mLatency tracing in distributed Java applications
46mPintrace: A Distributed Tracing Pipeline
34mDistributed Tracing, Lessons Learned
01h 11mDistributed Tracing : Latency Analysis for Your Microservices
41mTracing Real-Time Distributed Systems