A Verification-Constrained Framework for Safe Autonomous Remediation and Rollback in Distributed Systems

Abstract

The proposed approach in this research combines machine learning-based anomaly detection, safety evaluation and rollback mechanisms to create an autonomy-based framework for remediation of the distributed cloud systems which has verification constraints. It is used to assess the situation of the infrastructure, warn of abnormal conditions, assess operations, and assess recovery results. An evaluation in the laboratory confirms benefits of controlled autonomous responses in the context of resilience and reliability in cloud dynamic environments.