cncf.io
|
ksl
|
|
Kubernetes v1.35, tagged “Timbernetes,” ships workload-aware scheduling in alpha – gang scheduling that forces all-or-nothing Pod placement, which matters most for distributed training jobs that stall when partially placed. In-place Pod resize graduates to stable, letting teams adjust CPU and memory on inference services without restarting containers. Dynamic Resource Allocation continues maturing as the path to cleaner accelerator orchestration, and kubectl now defaults to KYAML, a stricter YAML subset meant to reduce configuration hazards at scale. Worth noting separately: Ingress NGINX enters end-of-life maintenance through March 2026 before full retirement. Kubernetes has been steadily absorbing AI infrastructure concerns release by release, and 1.35 makes the direction explicit rather than incidental.
