Formal Verification of Multi-Controller Systems in Cloud Management
Résumé
Kubernetes, a leading cloud orchestration system, relies on a diverse set of control components called controllers to manage resources through their reconciliation loops. A lack of coordination and communication between controllers can lead to conflicts between these reconciling loops. This problem actually stems from a novel computing paradigm in cloud management that has not yet been extensively studied by the academic community. To fill this gap, we first investigate specific controller conflict failures based on real-world case studies. Then, we provide a multi-controller model to abstract this new paradigm and propose convergence conditions that guarantee system convergence. We also construct a corresponding formal model, where convergence conditions are represented as LTL (Linear Temporal Logic) properties, and verify these properties using NuXMV. The verification results validate two main root causes of controller conflicts: resource overlap and resource dependency, which also reveal the inherent contradictions of the current multi-controller computing paradigm. Our study is an important step towards the construction of correct-by-construction cloud orchestration systems.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| licence |
