Amazon EKS adds advanced Kubernetes control-plane configuration

The new capability exposes previously managed control-plane knobs—kube-scheduler, kube-controller-manager, and API server settings—through EKS APIs, letting platform teams tune pod placement (e.g., MostAllocated bin-packing for denser scheduling), autoscaling responsiveness, and event retention without leaving the managed-EKS model. AWS published a walkthrough demonstrating the tuning.

Separately but in the same EKS 1.34 release, AWS made the Kubelet Checkpoint API functional, enabling forensic container checkpointing: capturing a running container's full state—memory, processes, and network connections—without stopping the workload, then storing checkpoints in Amazon ECR as OCI images via an unprivileged checkpoint agent. That's a meaningful capability for incident response and debugging live workloads.

Competitively, deeper control-plane configurability narrows the gap between managed EKS and self-managed Kubernetes, appealing to sophisticated platform teams weighing GKE and AKS. It fits AWS's week of infrastructure-depth announcements.

These are developer-infrastructure features rather than headline AI news, but they matter for teams running large agentic and inference workloads on Kubernetes. Skeptics will note that more configurability can reintroduce the operational complexity managed Kubernetes was meant to hide. Watch adoption among large EKS shops and whether misconfiguration incidents rise with the new knobs.