Holistic root cause analysis of software breakages through structural causal modeling
summary
Claims a fast, automated, holistic root cause analysis that uses an ensemble of structural causal models trained with the doy library and assigns contributions with Shapley values. On simulated microservice graphs the system pinpoints injected faults even when deep in the call chain, needs only two to five time steps post‑anomaly to start diagnosing, and produces a prototype in weeks with backtests showing promising accuracy, contingent on a reliable dependency graph and instrumentation.
AI-written summary · about
chapters (9)
- 0:00Observability and breakages
- 1:48What observability is
- 3:53Root cause analysis challenges
- 8:50Structural causal models
- 16:18Validating the approach
- 21:38Model assumptions and data
- 26:58Limitations and challenges
- 31:12Future plans and adoption
- 37:49Q&A
related talks
24mAutomated Root Cause Analysis
31mPost-mortems: Building better software together.
30mThere is No Root Cause...
20mAutoARTS: Taxonomy, Insights and Tools for Root Cause Labelling of Incidents in...
18mProduction software keeps breaking and it will only get worse
29m🧠🧹 Causality - Mental Hygiene for Data Science
24mHow Deep Systems Broke Observability
49mShifting Safety Techniques To a Statistical World