Kubernetes Multi-Tenancy Using Operators

Abstract

Over the past decade, Containers have taken center stage in software development due to their lightweight design and rapid deployment capabilities. This trend has led to the development of intricate applications, broken down into hundreds of containers, necessitating container orchestration systems. Kubernetes is a popular container orchestration platform that maintains the desired state of a container-based application. Large enterprise-scale Kubernetes clusters can simultaneously accommodate multiple users for their applications using soft or hard multi-tenancy approaches. Although this is resource-efficient, it can lead to noisy neighbor issues in an over-committed cluster. Moreover, it is challenging to utilize application-specific specialized hardware due to the hardware-agnostic nature of Kubernetes scheduler algorithms. It is extremely laborious for cluster administrations to provision and configure scheduling parameters on the specialized nodes based on application requirements. This thesis attempts to overcome these challenges by designing an operator: Dedicated Nodes operator. The operator enables the end-users to provision their nodes in the shared Kubernetes cluster without additional cluster-admin rights. It also ensures that the dedicated nodes are configured with the appropriate parameters to maintain node exclusivity for the user’s application. The effectiveness of the operator functionality is assessed and the impact of resulting dedicated nodes on the performance of the applications in a multi-tenant cluster is evaluated.