"This feature allows cluster operators to adjust CPU and memory resources for running pods without triggering a container restart. Piotr Mińkowski, a Solutions Architect at Red Hat, highlighted the importance of this to Java developers in his recent post on X.com. Why is it essential for Java? You can give the pod extra CPU only for startup, then shrink it after. App starts fast, the Pod uses the right amount of resources always."
"Alpha features in this release include native support for Gang Scheduling within the scheduler. Gang scheduling ensures that a group of interrelated pods, for example, AI/ML training jobs, are scheduled simultaneously or not at all. The new API resource in version 1.35 allows users to define scheduling requirements directly in the core API. In prior releases of Kubernetes, projects such as Volcano or Kueue were used to handle similar challenges."
Kubernetes 1.35, named "Timbernetes", emphasizes mutability and optimizes high-performance AI/ML workloads. In-Place Pod Resize reached general availability, allowing CPU and memory adjustments for running pods without container restarts. Java workloads can receive extra CPU for startup and be shrunk afterward to balance startup performance and steady-state resource usage. Alpha additions include native gang scheduling and a new API resource to express scheduling requirements in core Kubernetes, reducing the need for external projects like Volcano or Kueue. /flagz and /statusz endpoints were extended to produce machine-parsable output for authorized users. Configurable Horizontal Pod Autoscaler tolerance graduated to beta and became default.
Read at InfoQ
Unable to calculate read time
Collection
[
|
...
]