Kubernetes 1.37 Brings Stable Metrics API, Rootless Kubelet Beta, and 67 New Features

The Cloud Native Computing Foundation has unveiled Kubernetes 1.37—codenamed "Garhwal"—a release targeting operational stability, enhanced security posture, and optimized handling of artificial intelligence and machine learning applications.

Metrics and Autoscaling Reach Stability#

The Metrics API has graduated to general availability, providing standardized observability across cluster infrastructure. This stabilization enables foundational autoscaling capabilities through the Horizontal Pod Autoscaler (HPA) and Vertical Pod Autoscaler (VPA), as well as standard monitoring commands like kubectl top. Additionally, HPA scale-to-zero support—which transitioned to beta in version 1.27 and is now enabled by default—helps organizations reduce operational costs, particularly valuable for teams running GPU-intensive AI and ML workloads in cloud environments.

Enhanced Workload Management#

This release introduces several features aimed at improving how Kubernetes schedules and manages workloads:

• Workload-aware scheduling (alpha): The InPlacePodVerticalScalingSchedulerPreemption feature allows the scheduler to intelligently reclaim node resources by pausing lower-priority tasks, freeing capacity for in-place resizing of critical applications.
• StatefulSet Recreate strategy (alpha): A new deployment option for StatefulSets that cleanly removes all existing pods during updates, helping operators recover from stuck or pending states without manual intervention.
• Pod checkpoint and restore (alpha): The kubelet can now snapshot a running container's memory state and process tree, supporting debugging workflows and security investigations.

Security Advances#

Rootless Kubelet mode has advanced to beta, with the KubeletInUserNamespace feature gate now enabled by default. This capability allows core node components to operate as unprivileged users through Linux user namespaces, substantially reducing the blast radius of container escape vulnerabilities.

Another security milestone: reached stable status, offering native infrastructure for encrypted pod-to-pod communication. Organizations can now establish mutual TLS between internal services without external tooling.