Kubernetes v1.36: Advancing Workload-Aware Scheduling
Kubernetes v1.36 introduces the Workload and PodGroup APIs under scheduling.k8s.io/v1alpha2, separating the static Workload template from runtime PodGroup state. It adds a PodGroup scheduling cycle that evaluates a group atomically, along with first iterations of topology-aware scheduling and workload-aware preemption. The release also adds ResourceClaim support for PodGroups and a first phase of Job controller integration with the new API.
What changes for platform teams
Teams running batch or AI/ML training workloads on Kubernetes will see the v1alpha2 API replace v1alpha1, and the Pod workloadRef field is replaced by schedulingGroup, so manifests and controllers need updating. Gang scheduling and topology placement are alpha, and the source notes limits on heterogeneous groups and inter-Pod dependencies, so plan for partial guarantees.
Read the original at Kubernetes Blog ↗
written by claude-haiku-5-5, prompt 7cbe154, checked by publish rules