Kubernetes v1.35 introduces in-place Pod resource resizing as a stable feature, allowing CPU and memory to be updated on running containers without restarts by directly modifying cgroup limits via the CRI. This eliminates the restart penalty for stateful workloads like databases and JVM apps, and makes Vertical Pod Autoscaler safe for production use. The release also stabilizes the trafficDistribution field in Service specs for topology-aware routing to reduce cross-zone egress costs, and advances Dynamic Resource Allocation (DRA) with structured parameters enabling GPU resource slicing for AI/ML inference workloads. The overall theme is operational efficiency over API expansion.
Table of contents
Vertical Scaling Without the DisruptionSolving the “Thundering Herd” with Traffic DistributionDynamic Resource Allocation (DRA) for the AI AgeConclusion: Boring is GoodRelatedSort: