Kubernetes, the open-source container orchestration solution released by Google in 2014, has become a highly sought-after skill in recent years. It has evolved into a comprehensive ecosystem for managing microservices across cloud platforms. However, as AI/ML workloads became more prevalent, companies faced challenges in using Kubernetes due to its high resource requirements. This has led to some companies, such as Gitpod, deciding to shift away from Kubernetes and develop their own solutions. According to Gitpod’s co-founder and CTO, Christian Weichel, and staff engineer, Alejandro de Brito Fontes, while Kubernetes initially seemed like the obvious choice for their development platform, it proved to be ill-suited for unpredictable development environments. The team faced significant challenges with complexity, resource management, and state handling, leading them to develop their own solution called Flex. This sentiment is not unique to Gitpod, as many teams have underestimated the complexity of managing Kubernetes at scale. In response, the Cloud Native Computing Foundation (CNCF) released Kubernetes 1.31 (Elli) in August, specifically designed to improve resource management and efficiency for handling AI/ML applications. However, some experts, such as Murli Thirumale from Pure Storage, believe that Kubernetes may not be the best solution for managing AI/ML workloads and that companies should carefully consider their options before adopting it.
