Kubernetes 1.35, codenamed “Timbernetes,” brings significant advancements aimed at optimizing performance and resource management, particularly benefiting AI/ML workloads. This release introduces several key features that streamline operations and reduce costs associated with pod rescheduling and resource adjustments.
A notable addition to this version is the alpha feature known as Restart All Containers. It enables in-place pod restarts without the need for deletion and recreation, preserving the pod UID, IP address, volumes, and network namespace. This feature is especially beneficial when dealing with scenarios where inter-container dependencies necessitate a full reset. By eliminating the need to reschedule thousands of pods, especially in AI/ML workloads, the feature can help save over $100,000 monthly in compute resources. It operates by triggering container restart policy rules with the <code>RestartAllContainers</code> action, ensuring the full execution of the startup sequence, including init containers.
In addition to in-place pod restarts, Kubernetes 1.35 stabilizes the in-place pod resize capability, allowing for CPU and memory adjustments without restarting containers. This development is crucial for dynamic resource allocation, enabling more efficient scaling of applications based on current demand.
The version further enhances native gang scheduling, now in alpha, which orchestrates AI/ML training jobs more effectively by coordinating task execution across multiple pods. This is complemented by upgrades to the Horizontal Pod Autoscaler (HPA), which now supports configurable tolerance thresholds, promoting more precise scaling decisions.
Kubernetes 1.35 also sees the promotion of PodCertificateRequests to beta, easing the management of mTLS certificates by simplifying and automating certificate requests.
As part of its ongoing evolution, the Kubernetes community has announced the planned retirement of the Ingress NGINX controller, set for March 2026, in anticipation of shifting towards the newer Gateway API.
Overall, “Timbernetes” with its 60 enhancements, marks a significant stride in Kubernetes’ development, reinforcing its position as a preferred platform for efficient and scalable application orchestration.

Collections

Kubernetes 1.35 (Timbernetes) introduces in-place pod restarts that preserve pod identity and network configuration, potentially saving significant compute costs for AI/ML workloads. The release stabilizes in-place pod resizing for CPU and memory adjustments, adds alpha support for native gang scheduling to coordinate multi-pod training jobs, and enhances the Horizontal Pod Autoscaler with configurable tolerance thresholds. PodCertificateRequests reach beta for simplified mTLS management, while the community announces the planned retirement of Ingress NGINX controller in favor of Gateway API by March 2026.

Kubernetes 1.35 Enhances Efficiency with In-Place Pod Restart and AI-Optimized Features