Kubernetes v1.36 enhances security and AI support

Kubernetes v1.36 enhances security and AI support

The Kubernetes project has released version 1.36 with 71 improvements, comprising 18 stable features, 26 beta features, and 25 alpha features. The release focuses primarily on improved access control, visibility into hardware failures, and native support for complex AI and batch workloads.

One of the most notable additions is fine-grained kubelet API authorization, which is now generally available. With least-privileged access control for the kubelet HTTPS API, users no longer depend on node/proxy permissions for monitoring and observability tasks. Cluster operators can now grant specific access to individual kubelet endpoints. This should increase a cluster’s security level, as only the permissions necessary for operation are granted.

Additionally, Resource Health Status has reached beta. Kubernetes previously lacked a native way to report the health of allocated hardware, making it difficult to link pod crashes to faulty hardware. Version 1.36 extends this feature to Dynamic Resource Allocation (DRA) via the `allocatedResourcesStatus` field. Administrators can now use `kubectl describe pod` to see if a container crash is related to an `Unhealthy` or `Unknown` device status.

Gang Scheduling for AI and Batch Processing

As an alpha feature, v1.36 introduces a suite of new Workload Aware Scheduling (WAS) features. Until now, the Kubernetes scheduler and job controllers treated pods as separate units, leading to resource waste in distributed workloads. With the new Workload API and PodGroup API, related pods are treated as a single logical unit.

This allows the scheduler to perform Gang Scheduling: pods are only bound together once the required minimum number is ready. Topology-Aware and Preemption policies also place pods optimally within network or rack domains. This reduces the need for third-party schedulers in AI/ML and batch environments. The functionality has been developed in several Kubernetes Enhancement Proposals led by SIG Scheduling and SIG Apps.

Tip: More flexible Kubernetes management with Cluster API 1.12