Automated Root Cause Analysis
summary
Swisscom builds Mosaic: a distributed‑tracing pipeline that ingests spans via Tempo and Kafka, converts span chains into functions with three metrics (calls, errors, P95), trains an ML forecaster on two months of history to predict 24‑hour bounds and detect anomalies (alerts now in 30s–1min), aggregates into a health score 0–100%, and assembles causal graphs from anomalies, events and dependencies. A similarity module (weighted attributes + sigmoid) ranks past incidents and a RAG and LLM step reads investigators’ comments to boost probability estimates and point to likely root causes (speeds RCA; not fully automatic).
AI-written summary · about
chapters (9)
- 0:00Introduction
- 2:59Mosaic platform and tracing
- 8:12Tracing to anomaly detection
- 10:16Automated RCA definition
- 13:48Building causal graphs
- 16:15Past incidents for RCA
- 18:23AI-assisted investigations
- 20:29Live RCA demo
- 23:21Q&A
related talks
32mGenerative AI for Observability: Automating Root Cause Analysis in Modern IT Systems
31mFrom Alerts to Insights: Performing Trace-Based Causation at Scale
38mHolistic root cause analysis of software breakages through structural causal modeling
14mCognitive and SelfAdaptive System for Distributed Tracing
27mAI-Powered Root Cause Analysis at Scale: From Theory To Production...
45mTracing: A journey to tactical insights
32mA Picture is Worth 1,000 Traces
30mDebugging at Scale Using Elastic and Machine Learning