Kubernetes v1.35 Introduces Native Workload Aware Scheduling
The release of Kubernetes version 1.35 marks a significant advancement in cluster management by introducing native support for workload aware scheduling. This update addresses the complexities of scheduling large, multi-Pod applications, such as machine learning batch jobs, which require strategic placement for efficiency. Previously reliant on custom schedulers, users can now utilize the new Workload API within the scheduling.k8s.io/v1alpha1 group to define scheduling requirements directly. Key features include an initial implementation of gang scheduling, which ensures Pods are scheduled in an all-or-nothing fashion, and opportunistic batching to accelerate the handling of identical Pods. This development aims to make workloads first-class citizens in the kube-scheduler, reflecting the growing importance of AI use cases. The initiative is part of a broader, multi-release effort involving various Special Interest Groups (SIGs) to enhance workload management capabilities, including future improvements in preemption and autoscaling. By providing a structured, machine-readable definition for multi-Pod applications, Kubernetes seeks to streamline the lifecycle management of complex workloads, reducing fragility and improving overall system performance for data-intensive tasks.
Wire timeline
Kubernetes v1.35 Introduces Native Workload Aware Scheduling
The release of Kubernetes version 1.35 marks a significant advancement in cluster management by introducing native support for workload aware scheduling. This update addresses the complexities of scheduling large, multi-Pod applications, such as machine learning batch jobs, which require strategic placement for efficiency. Previously reliant on custom schedulers, users can now utilize the new Workload API within the scheduling.k8s.io/v1alpha1 group to define scheduling requirements directly. Key features include an initial implementation of gang scheduling, which ensures Pods are scheduled in an all-or-nothing fashion, and opportunistic batching to accelerate the handling of identical Pods. This development aims to make workloads first-class citizens in the kube-scheduler, reflecting the growing importance of AI use cases. The initiative is part of a broader, multi-release effort involving various Special Interest Groups (SIGs) to enhance workload management capabilities, including future improvements in preemption and autoscaling. By providing a structured, machine-readable definition for multi-Pod applications, Kubernetes seeks to streamline the lifecycle management of complex workloads, reducing fragility and improving overall system performance for data-intensive tasks.
Kubernetes Blog