Who Watches the Watchers? On the Reliability of Softwarizing Cloud Application Management
summary
Study analyzes 412 real-world failures across 13 Kubernetes operators and finds 42% stem from operator–application interactions, with 64% of those being semantic violations. A push-button tester called Oat synthesizes application-specific states, injects application-targeted faults, and uncovers 86 new bugs in six popular operators. The evaluation shows 72% of failures need application-specific properties and 41% require injected faults, highlighting lack of a well-specified management interface.
AI-written summary · about
chapters (8)
- 0:00Operator reliability issues
- 2:05Operators in production failures
- 4:30412 real-world operator failures
- 6:58Failure patterns in apps
- 9:26Semantic violations in logic
- 11:34Oat testing tool
- 14:06Oat evaluation results
- 15:29Future of management interfaces
related talks
24mKubernetes Operators: Safety First Through Model Checkers
27mRethinking Observability for Off-the-Shelf Tools
43mKubernetes Operability Tooling
40mThe Good, the Bad, and the Ugly of Managing Production Systems with Kubernetes
35mHow to Cloud Ops Like a Boss
31mLow-Level Computing in a Cloud Native World
42mMastering Kubernetes: Challenges, Strategies, and Best Practices
58mObserving Applications in Production