Scheduling - 1

2026 (2)

January (2)

k8s: Announcing the Checkpoint/Restore Working Group

Kubernetes Official Blog

2026-01-21

Kubernetes has launched the Checkpoint Restore Working Group to advance integration of checkpoint/restore functionality (primarily via the CRIU ecosystem) into Kubernetes. The WG will explore use cases like faster startup, resource optimization for interactive AI workloads, fault-tolerance for long runs, interruption-aware scheduling, live Pod migration, and forensic checkpointing, and invites contributors to join meetings, Slack, and the mailing list.

k8s: Kubernetes v1.35: Extended Toleration Operators to Support Numeric Comparisons (Alpha)

Kubernetes Official Blog

2026-01-05

Kubernetes v1.35 introduces Extended Toleration Operators (alpha) that add numeric comparison operators Gt and Lt to spec.tolerations. This lets tolerations compare numeric taint values (e.g., failure-probability, gpu-compute-score, disk-iops, cost-per-hour) so schedulers can make threshold-based placement decisions while preserving taint/toleration semantics (centralized node-side policy and eviction via NoExecute). The feature is gated, supports all taint effects, has specific numeric format rules, and is experimental — feedback and further enhancements (CEL, autoscaling integration, graduation) are planned.

2025 (2)

December (1)

k8s: Kubernetes v1.35: Introducing Workload Aware Scheduling

Kubernetes Official Blog

2025-12-29

Kubernetes v1.35 introduces workload-aware scheduling with a new Workload API (scheduling.k8s.io/v1alpha1) to describe multi-Pod scheduling requirements, an initial gang scheduling implementation for all-or-nothing placement, and opportunistic batching (Beta) to speed scheduling of identical Pods. Gang scheduling uses podGroups, a workloadRef on Pods, and a Permit gate with a 5-minute timeout; opportunistic batching reuses feasibility checks for identical Pods but has strict identical-field restrictions. The release also outlines a broader roadmap (workload-level preemption, autoscaling integration, topology-aware scheduling) and explains required feature gates and how to test and give feedback.

September (1)

k8s: Kubernetes v1.34: Pod Level Resources Graduated to Beta

Kubernetes Official Blog

2025-09-22

Kubernetes v1.34 graduates Pod Level Resources to Beta and enables it by default. The feature lets you declare CPU, memory and hugepages for an entire Pod (in addition to per-container settings), with pod-level requests used for scheduling and pod-level limits acting as an absolute runtime cap; it improves intra-pod sharing and QoS handling but has several platform and feature limitations.