← back

📷 "Technology Abstract" by Lenharth Systems is marked with CC0 1.0. To view the terms, visit https://creativecommons.org/publicdomain/zero/1.0/.

Kubernetes v1.36 'Haru': Enhanced Security and AI Focus for Container Orchestration

19 July 2026 · 4 min · Martin Jochum #Kubernetes#Container#KI#Sicherheit#DevOps

On April 22, 2026, Kubernetes v1.36 was released under the codename «Haru» (ハル) — the first major Kubernetes version of the year. With 70 enhancements (18 Stable, 25 Beta, 25 Alpha), the release sets new accents in the areas of security, AI workload management, and API scalability. 106 companies and 491 individuals contributed to it.

User Namespaces finally production-ready

The most important security innovation is the GA status of User Namespaces (KEP #127). This long-awaited feature allows mapping a container’s root user to a non-privileged user on the host system. Even if a process escapes from a container, it does not gain administrative rights on the underlying node. This closes one of the biggest security gaps in multi-tenant environments and provides effective protection against container breakout attacks.

Also promoted to stable status are Mutating Admission Policies (KEP #3962). Administrators can now define mutations of API resources directly in the API server using the Common Expression Language (CEL) — without needing to operate a separate webhook server. This saves latency and significantly reduces operational complexity.

More granular access control and security

Fine-Grained Kubelet API Authorization (KEP #2862) also reaches GA. Instead of the previous coarse nodes/proxy permission, monitoring and observability tools can now obtain precisely configured accesses to the kubelet HTTPS API following the least-privilege principle. This is a big step for security-conscious cluster operators.

With SELinux Volume Labeling (GA), recursive file relabeling is replaced by a mount -o context=XYZ option. This significantly speeds up pod startup on SELinux-based systems.

AI Workloads: New foundations for GPU clusters

Dynamic Resource Allocation (DRA) receives several important improvements in v1.36. DRA Admin Access and Prioritized Lists reach GA and provide a stable framework for managing hardware resources such as GPUs across cluster boundaries. DRA Partitionable Devices, Consumable Capacity, and Device Taints and Tolerations move to Beta status and are enabled by default.

A new Alpha feature is Workload-Aware Preemption: Previously, the scheduler could terminate individual pods early to make room for higher-priority workloads — a problem for distributed training where seven out of eight ranks are running but the eighth is missing. The new feature treats a PodGroup as a single preemption unit and only performs eviction when the entire group can be placed.

The Gang Scheduling API moves from Alpha to Beta. Mutable Pod Resources for Suspended Jobs (Beta) allows adjusting CPU, memory, and GPU requirements for running jobs without restarting pods — a gain for queue controllers in AI environments.

API scalability and infrastructure

A new Alpha feature is Sharded List and Watch Streams: In large clusters with many controllers, watch streams can become bottlenecks. The new architecture distributes the load across multiple streams.

Volume Group Snapshots (GA) enable crash-consistent snapshots across multiple PersistentVolumeClaims — essential for databases and StatefulSets. OCI Volume Sources (GA) allow mounting container images and artifacts directly from OCI registries as volumes.

Pressure Stall Information (PSI) (GA) delivers precise metrics on CPU, memory, and I/O bottlenecks based on cgroupv2 and helps detect «noisy neighbor» effects early.

Important removals and changes

When upgrading to v1.36, the following removals should be noted:

  • The gitRepo volume plugin (deprecated since v1.11) has been removed completely.
  • The IPVS mode in kube-proxy (deprecated since v1.35) is no longer available.
  • FlexVolume support in kubeadm and the Portworx in-tree driver have been removed.

Also worth mentioning: The Ingress NGINX project was discontinued on March 24, 2026 by SIG Network and the Security Response Committee. Since then, there have been no releases, bug fixes, or security updates. Migrations to Gateway API or alternatives should be prioritized.

Conclusion

Kubernetes v1.36 «Haru» is a solid release that elevates security fundamentals to production level while laying the foundation for the next generation of AI workloads. VMware Cloud Foundation aptly describes the trend: «Kubernetes is moving from a flexible framework toward more opinionated defaults for security and resource management.» The focus on User Namespaces, DRA, and Workload-Aware Scheduling shows that the orchestration platform shows no signs of slowing innovation even after more than ten years.

Sources

🌐 Machine-translated from the German original, editorially reviewed. 🤖 Written with AI assistance.