Kubernetes v1.37 introduces scheduler preemption for in-place Pod resizing to resolve the 'Deferred' status, which previously left scale-up requests blocked on fully utilized nodes. By enabling the InPlacePodVerticalScalingSchedulerPreemption feature gate, the scheduler can now preempt lower-priority pods to free up capacity, allowing higher-priority workloads to resize without application restarts.
Key points
In-place Pod resizing can move to 'Deferred' status if a node lacks sufficient headroom for a requested resource increase.
The new InPlacePodVerticalScalingSchedulerPreemption feature allows the scheduler to proactively evict lower-priority pods to accommodate higher-priority resize requests.
This mechanism prevents disruptive pod rescheduling by allowing resources to be adjusted dynamically on the current host node.
Preemption is only triggered when resources are insufficient, maintaining the core value proposition of non-disruptive scaling.