In Kubernetes cluster management, updating worker pools often involves rolling replacement—creating new nodes with updated configurations and deprecating old ones. However, this process poses challenges in certain environments. Physical machines face challenges like long boot times, slow cleanup/sanitization, local storage dependencies, and limited interchangeability. Similarly, virtual machines encounter resource shortages needed for rolling updates during capacity crunches or when scarce resources like GPUs are attached.
To address these challenges, we propose in-place update strategies for machine image and Kubernetes minor version updates, as well as certificate rotations. This approach minimizes resource waste, reduces downtime, and simplifies operations. This talk will explore these strategies and demonstrate their implementation using Gardener’s in-house OS, Garden Linux, and the machine-controller-manager component which orchestrates the lifecycle of VMs across infrastructures.