Automated Canary Analysis for Kubernetes Deployments

Automated Canary Analysis for Kubernetes Deployments

연구 분야: Software Development

학회: 2025 IEEE Cloud Summit

When organizations start implementing microservices and the concept of continuous delivery pipeline, especially while using DevOps, deploying becomes a crucial task in which safety is a significant concern. Canary releases have become a profound practice for rolling out new application updates to a limited number of users in an organization to avoid introducing new bugs in the critical production environment. Nonetheless, catching and validating issues during canary rollouts is often an irreversible process prone to mistakes. This technical paper focuses on the concept of Automated Canary Analysis for the implementation of Kubernetes and the role performed by the ACA over Service Mesh, Observability, and Metrics for making efficient deployment decisions. ACA derives performance differences as well as user experience effects between canary and baseline versions through statistical and machine learning algorithms. The study also considers existing open-source tools, namely Kayenta and Flagger, and establishes their capacity to perform automation of progressive delivery. Our results show a reduction in the use of the deployment risk factor while also improving the efficiency of the developers and the overall functionality of the system; therefore, ACA should be considered an important element of modern DevOps.