Rex: Preventing Bugs and Misconfiguration in Large Services Using Correlated
summary
Rex learns correlated change‑rules from version histories using association rule mining and then refines them with tree-based diffs and anti‑unification to identify the specific edits that trigger partner-file changes. It runs as a suggestion engine at PR time, processes >1,000 PRs/day, and is deployed 14 months across 360 repositories (Office 365, Azure). It surfaces ~5,000 developer-acted recommendations, yields about 20% deployment precision after refinement, and prevents real misconfigurations that would otherwise cause service disruption.
AI-written summary · about
chapters (16)
- 0:00Intro
- 0:20Scenario : Renaming A Microservice
- 1:35Why Do We Need Configuration Files?
- 2:26Why Is Configuration Management Difficult?
- 5:181. Change Rule Discovery - Model
- 6:27Rex: Correlated Change Analysis Tool
- 8:222. Change Rule Refinement
- 9:23Learning Generalizations
- 10:32Stage 1: Generate Diff Data
- 11:25Stage 2: Generalization
- 12:24Hierarchical Clustering
- 13:19Enhanced Rules
- 13:49Deployment
- 14:32Evaluation Metrics
- 16:20User Study
- 17:44Summary
related talks
44mREX of a multi-agency, multi-BU, multi-tech Cloud Native platform
47mREX QuickSign/Integrating security into a DevOps environment: Mission Impossible?
18mRed Tide Revert
44mRegression tracking & fixing: current state, problems, and next steps
44mRex Scale-up: The Impacts of Scaling Up
27mCatch Me If You Can: Hunting Misconfigurations Before They...
23mFrog-Proof Security: Streamlining The Sec In DevSecOps
30mThe Path to SRE