Diffy: Data-Driven Bug Finding for Configurations
summary
Diffy automatically synthesizes structured configuration templates with holes, then flags anomalous template parameters using unsupervised anomaly detection. Implementation combines a dynamic-programming string-pattern learner and sequence/list clustering, runs on millions of lines and finishes within 20 minutes, and is about two orders of magnitude faster than prior string-synthesizers. Evaluation finds 62 true positives on a 5G dataset, recovers 74% of prior MySQL issues with far fewer false positives, and catches real outages (a 1200s timer misconfiguration) using template synthesis + isolation forest.
AI-written summary · about
chapters (6)
- 0:00Config bug finding intro
- 4:12Diffy config template synthesis
- 8:27String template algorithm
- 13:07Anomalies with Isolation Forest
- 14:10Diffy performance on datasets
- 17:25Q&A
related talks
38mDIFFY Catching Bugs without Writing Tests
38mDiffy - Predicting and Preempting Regressions in Services - Puneet Khanduri & Prashant Khanduri
06mUnderstanding Configuration Issues in Storage Systems (Work-in-Progress Report)
32mConfiguration Is (riskier than?) Code
27mCatch Me If You Can: Hunting Misconfigurations Before They...
19mConfig as Code
29mData Diffing Based Software Architecture Patterns
34mThe Field Guide to Understanding Declarative Systems