Back to Blog
Cloud & Infrastructure April 24, 2026 5 min read

Kubernetes v1.36 'Haru' Ships — Rootless Containers Now Native, HPA Scales to Zero

Kubernetes v1.36, codenamed Haru, landed on April 22 with 70 enhancements. Rootless containers via user namespaces are now stable and production-ready, MutatingAdmissionPolicy is GA, and HPA can now scale deployments all the way to zero replicas.

Kubernetes v1.36 'Haru' Ships — Rootless Containers Now Native, HPA Scales to Zero

Kubernetes v1.36 “Haru” shipped on April 22, 2026. Seventy enhancements made the cut: 18 graduating to stable, 25 entering beta, 25 to alpha. The headline features are the ones that remove long-standing gaps — rootless containers done natively, mutation webhooks replaced by native objects, and HPA finally reaching zero.

User Namespaces are stable

Running rootless containers in Kubernetes used to mean stitching together third-party tooling. In v1.36, user namespaces are native, stable, and production-ready. Enable it with hostUsers: false in your pod spec. Containers run as a non-root user on the host even when they appear as root inside the container. This closes a meaningful attack surface and is especially relevant for multi-tenant clusters where workload isolation matters.

MutatingAdmissionPolicy is GA

Mutation webhooks are now a legacy pattern. MutatingAdmissionPolicy lets you express mutation logic as native Kubernetes objects — no external service, no webhook latency, no additional failure domain. It uses the same CEL expression language as ValidatingAdmissionPolicy, which reached GA in 1.30. If your cluster is still running mutation webhooks for standard policy enforcement, the migration path is clear.

HPA scale-to-zero

The Horizontal Pod Autoscaler can now scale a deployment down to zero replicas when there’s no workload and back up on demand. Combined with KEDA or custom metrics, this unlocks a genuinely serverless execution model within Kubernetes. For batch workloads and event-driven jobs that sit idle most of the day, the compute savings are immediate.

OCI volumes are stable

OCI VolumeSource is now generally available — a new volume type that mounts OCI images directly into pods. The use case: ship static assets, configuration bundles, or model weights as container images and mount them at runtime without baking them into the application image. Useful for AI inference workloads where model files change independently of application code.

What’s gone

The gitRepo volume type is removed — it was disabled since v1.11 and is now completely unrevivable. Plan accordingly if anything in your cluster still references it. The externalIPs field in Service specs is being deprecated, with removal scheduled for v1.43. Most significantly: Ingress-NGINX was retired by SIG Security on March 24, 2026. No further security patches will be published. If you’re running Ingress-NGINX in production, migrate to an alternative ingress controller now — not when the next CVE drops.

The SELinux improvement

SELinux volume mounting is GA. The old behavior relabeled files recursively at container start — slow and sometimes problematic on large volumes. The new behavior applies mount -o context=XYZ at mount time, which is atomic and fast. Pods on SELinux-enforcing systems should see noticeably faster startup.

v1.37 is already in planning. The momentum on AI workload primitives — GPU resource partitioning, DRA improvements, and topology-aware scheduling — suggests the next release will go deeper on the infrastructure layer Kubernetes needs to own for model serving.

Kubernetes cloud containers infrastructure open source