KEP-5836: Scheduler Preemption for In-Place Pod Resize

Implementation History
ALPHA Implementable
Created 2026-02-23
Latest v1.37
Milestones
Alpha v1.37
Beta v1.38
Stable v1.39
Related Enhancements
Ownership
Owning SIG
SIG Scheduling
Participating SIGs
SIG NodeSIG Autoscaling
Primary Authors
@natasha41575

KEP-5836: Scheduler Preemption for In-Place Pod Resize

Release Signoff Checklist

Items marked with (R) are required prior to targeting to a milestone / release.

  • (R) Enhancement issue in release milestone, which links to KEP dir in kubernetes/enhancements (not the initial KEP PR)
  • (R) KEP approvers have approved the KEP status as implementable
  • (R) Design details are appropriately documented
  • (R) Test plan is in place, giving consideration to SIG Architecture and SIG Testing input (including test refactors)
    • e2e Tests for all Beta API Operations (endpoints)
    • (R) Ensure GA e2e tests meet requirements for Conformance Tests
    • (R) Minimum Two Week Window for GA e2e tests to prove flake free
  • (R) Graduation criteria is in place
  • (R) Production readiness review completed
  • (R) Production readiness review approved
  • “Implementation History” section is up-to-date for milestone
  • User-facing documentation has been created in kubernetes/website , for publication to kubernetes.io
  • Supporting documentation—e.g., additional design documents, links to mailing list discussions/SIG meetings, relevant PRs/issues, release notes

Summary

In-Place Pod Resize reached GA in 1.35, allowing pod resizing without restarts. However, unlike traditional pod recreation, in-place resizes that exceed node capacity do not currently trigger the preemption of lower-priority pods. Instead, they remain in a Deferred resize status and require manual intervention. We propose that the Scheduler actively creates space for resize requests by preempting lower-priority pods, following the same preemption logic used when scheduling new pods.

Motivation

The introduction of In-Place Pod Resize gives platform teams new ways to optimize cloud resources without restarting pods, marking a significant shift in how Kubernetes operates. This is especially powerful when In-Place Pod Resize is integrated with higher-level autoscaling controllers such as VPA’s new InPlaceOrRecreate mode. However, users currently lack a way to control scale-up behavior when a node lacks the capacity to fulfill the request. Consequently, workloads may face disruptions such as being moved to a larger node, or suffering an OOM-kill because memory could not be scaled up in-place.

Scheduler preemption eliminates the gap by introducing an configurable ability to free up capacity on a fully-utilized node to allow the scale up to succeed in-place.

Goals

  1. Introduce the ability to instruct the Scheduler to preempt lower-priority pods to make room for a Deferred resize.
  2. Keep the preemption behavior for Deferred resizes aligned with the preemption behavior for newly created pods.

Non-Goals

  1. Perform the entire scheduling cycle (including binding, DRA, etc) on Deferred resizes; the Scheduler should trigger preemption and nothing more.
  2. Coordinate with higher-level autoscalers such as VPA or CA on the preemption decision. This is considered out of scope for this KEP, but may be considered as a future enhancement.
  3. Changing the source of truth for resize allocations. The Kubelet remains responsible for evaluating incoming resize requests, and the scheduler must account for cases where the Kubelet’s runtime decisions differ from the scheduler’s calculated predictions.

Proposal

In-Place Pod Resize requests that result in a Deferred resize status trigger scheduler preemption, keeping the behavior aligned with scheduler preemption for newly created pods.

The Scheduler watches for Deferred resizes. To make room for the resize, it tries to preempt (evict) lower-priority pods to make the resize possible. The Kubelet today watches for pod removals and reacts to the eviction by reattempting the previously Deferred resize.

User Stories (Optional)

Story 1: Pods in the Deferred resize status no longer require manual interversion.

Users are enabled to instruct the system to automatically evict lower-priority workloads rather than the resize request sitting there indefinitely.

Story 2: Reduction of disruption for critical workloads.

The Scheduler would preferentially evict lower-priority pods before an external controller, such as VPA’s InPlaceOrRecreate mode , would be forced to fall back to the more disruptive action of evicting and recreating the critical pod itself.

Story 3: Driving cluster autoscaling

If a pod managed by a workload controller is preempted, the corresponding controller will likely create new pods in response to the preemption, which will appear in the scheduling queue. By moving the resource deficit from an indefinite Deferred status into the active scheduling queue, preempted pods will naturally trigger cluster autoscalers to provision new capacity.

Risks and Mitigations

Performance impact

If there are a significant number of deferred pods, periodic processing of those pods can affect scheduling throughput. To mitigate this, deferred pods are only reconsidered if something on the node has changed that could make preemption succeed.

Interaction with workload-aware preemption

With workload-aware preemption, the resize may end up triggering preemption of a potentially large workload. In these cases, the preempted workload would always be lower priority than the pod that is being resized, so we consider this working as intended.

Race between a Deferred resize and a new higher-priority pod

From the scheduler’s view, once the spec is updated, the resources are already reserved. It doesn’t matter from the scheduler perspective whether the resize has been actuated by the Kubelet yet.

This means that if a new, higher-priority pod comes in and the only way to fit it is by taking the space the Deferred pod is trying to grow into, the standard preemption logic applies. This might mean the resizing pod itself gets evicted if it’s the best victim candidate. While we would rather not kill pods unnecessarily, this behavior is consistent with the rest of the scheduler’s logic.

Shifting Preemption Victims during Scheduler Restart

If a Deferred resize was mid-preemption when the scheduler crashed or restarted, the new scheduler instance might select a different victim pod than the original instance did. This can happen due to:

  • Changes in cluster state (new pods, node updates) during the scheduler’s downtime.
  • Non-deterministic tie-breaking when multiple low-priority pods satisfy the resource requirement equally well.

In specific edge cases, this leads to redundant preemption. “Victim A” (targeted by the first scheduler) and “Victim B” (targeted by the second scheduler) may both be terminated to satisfy a single resize request.

Mitigations:

  • Idempotency: The Delete API is already idempotent. If the scheduler picks the same victim upon restart, the API server simply acknowledges the request without further disruption.
  • Acceptable Waste: In the Kubernetes priority model, ensuring the success of a higher-priority workload (the resizing pod) justifies the potential loss of multiple lower-priority victims during a rare control-plane failure.

To address double preemption risks across scheduler restarts and long graceful termination periods, we are exploring tracking mechanisms (such as using the pod’s nominated node name). We will fully evaluate these options and finalize the approach for Beta; see details in Tracking Preemption Nominations to Avoid Double Preemption .

Risk of Additional Preemption

The Scheduler (via the NodeResourcesFit plugin) assumes the resources of pods to be max(desired, allocated, actual). This ensures that the node does not end up overcommitted in the event that Kubelet actuates a resize while the new pod is being scheduled. This behavior is correct for scheduling of newly created pods.

However, this logic results in potentially unnecessary preemption during scheduler-evaluation of Deferred pods.

The Kubelet determines resource fit in the event of a resize by assuming resources as:

  • For the pod that is being resized, use max(desired, allocated, actual).
  • For all other pods on the node, use max(allocated, actual) (ignoring desired).

Consider this case with 4 pods, all Deferred:

  • Pod1 (high priority): desired=X, allocated=X/2, actuated=X/2
  • Pod2,3,4 (low priority): desired=2X, allocated=X/2, actuated=X/2
  • Node allocatable=2X

If we use max(desired, allocated, actual) for all pods, then:

  1. Preemption logic first removes all candidate pods, so it sees node is using X (Pod 1’s max).
  2. The Scheduler first tries to reprieve pod 2 which is assumed to have max(2X, X/2, X/2) resources. 2X + X = 3X, which is more than node allocatable 2X. The Scheduler determines this pod cannot be reprieved.
  3. The Scheduler goes through the same process for pods 3 and 4, and determines that they cannot be reprieved.

All 3 pods (Pod 2, 3, and 4) would be preempted, even though it was only necessary to preempt one of them.

To prevent unnecessary preemption, the NodeResourcesFit plugin would need to likewise assume resources in the same way as the Kubelet when evaluating deferred pods, while maintaining its existing behavior for scheduling new pods.

If we use our modified logic, where only the resizing pod uses max(desired, allocated, actual) and the remaining pods use max(allocated, actual), then:

  1. Preemption logic first removes all candidate pods, so it sees node is using X (Pod 1’s max).
  2. The Scheduler first tries to reprieve pod 2, which assumed to have max(X/2, X/2) resources. X/2 + X = 1.5X, which is less than node allocatable 2X, so pod 2 is reprieved.
  3. The Scheduler tries to reprieve pod 3, which assumes to have max(X/2, X/2) resources. X/2 + 1.5X = 2X, which still fits within the node allocatable 2X, so pod 3 is reprieved.
  4. The Scheduler tries to reprieve pod 4, which assumes to have max(X/2, X/2) resources. X/2 + 2X = 2.5X, which is more than node allocatable 2X, so pod 4 is not reprieved.

This results in only 1 pod (Pod 4) being preempted.

Such a modification to the NodeResourceFit plugin is out of scope for alpha due to its wide ranging implications across the scheduler. However, we will reconsider this decision prior to beta.

Design Details

How Deferred Resizes Integrate into the Scheduling Queue

The scheduling queue is shared between new pods and Deferred pods, using existing queuing logic to sort them by their current scheduling priority.

See alternative considerations for separate scheduling queues here and prioritized resize logic here .

Detecting Deferred Resizes in UpdatePod

The UpdatePod event handler is responsible for maintaining the state of the scheduling queue based on changes to the pod objects, and includes logic to handle pods with a Deferred resize:

  • If the deferred condition is added to a pod: The Scheduler adds the pod to the scheduling queue.
  • If the deferred condition is removed: The Scheduler removes the pod from whatever queue it’s currently in. This can occur if the user reverts the resize request or if the Kubelet is able to satisfy the resize request due to other pods being removed or downsized.

Queueing Hints in NodeResourcesFit

If the Deferred pod’s spec (desired) resources change (which can happen if the desired resources are updated before the previously requested resize can complete), the Scheduler updates its view of the pod by leveraging NodeResourcesFit queueing hints (QHints). The NodeResourcesFit QHint for target pod updates detects this change and automatically moves the pod back to the active queue if resource changes are detected.

Scheduler Restart and State Recovery

Because the scheduler’s internal activeQ and Unschedulable pods pool are maintained in-memory, a scheduler restart clears the state of all pods currently undergoing preemption for a resize. To prevent these pods from remaining in a Deferred state indefinitely, the scheduler must proactively re-identify them upon startup.

  • Cold-Start Re-Queueing: As part of initialization, the scheduler adds deferred pods to the scheduling queue in the AddPod event handler.
  • Re-evaluation: Once in the queue, these pods undergo the standard evaluation flow (Node Fit -> Preemption).

See risk considerations for shifting victims here and alternative considerations for double preemption mitigation here .

Conflicting Sources of Truth for the Pod

Adding a snapshot of a deferred pod directly to the scheduling queue means the pod exists in two places simultaneously: the scheduling cache and the scheduling queue. Any subsequent pod updates (such as resize requests, priority updates, or label changes) must be propagated to both the cache and the queue in parallel.

The primary risk with this approach is ensuring that no existing or future scheduler logic implicitly assumes a pod has a single unique representation or reference. We must audit the scheduler codebase to ensure that:

  1. Pod updates correctly target both locations. The pod should be updated in the cache first, then in the queue, to make any divergences between the two more predictable.
  2. No downstream scheduling or preemption logic relies on strict pointer equality of pod objects across the cache and queue boundaries.
  3. Resource double-counting is avoided on the target node by excluding the pod’s additional resources during resource fit calculations as described in the section below.

See alternative considerations for how to handle this without dual representation (such as using a dynamic PodGetter interface) here .

Processing Deferred Resizes in the Scheduling Cycle

When processing a pod with a Deferred resize, the scheduler runs it through a modified scheduling cycle. Rather than evaluating the pod as if it were brand-new to the cluster, the cycle is restricted exclusively to resource capacity checks and preemption logic. This differs from standard scheduling in two key ways:

  • Targeted Node Search: Candidate node evaluation is restricted solely to the node the deferred pod is already running on.
  • Bypassing Irrelevant Constraints: All scheduling plugins irrelevant to direct resource capacity fitting (such as node affinity, pod anti-affinity, and topology spread constraints) are bypassed.

To achieve this cleanly and without hardcoding specific plugin names on the framework side, we adjust all irrelevant plugins to be skipped when processing Deferred pods.

Only NodeName, NodeResourcesFit, and DefaultPreemption plugins implement handling for processing Deferred pods. All other plugins are modified to implement the PreFilter interface and return Skip for Deferred pods, which results in the plugin being bypassed for deferred resize scheduling cycles. Out-of-tree plugin developers are advised to align their implementation for handling of Deferred pods.

NodeName Plugin: Node Restriction via the PreFilter Phase

The NodeName plugin is adjusted to implement the PreFilter phase. Because the deferred resize pod is already bound to and running on a host, the scheduler limits the candidate search space at the start of the scheduling cycle. If a pod specifies Spec.NodeName, its PreFilter returns a PreFilterResult containing only that node. This restricts the candidate node evaluation search space to just the pod’s currently assigned node.

NodeResourcesFit Plugin: Calculating Resource Fit in the Filter Phase

The NodeResourcesFit plugin will run its PreFilter and Filter phases for deferred resize pods. It runs a modified resource-fit check to specially handle Deferred pods.

To prevent the double-counting of resources when a deferred-resize pod is evaluated for scheduling or preemption fit on its assigned node, the NodeResourcesFit plugin is adjusted to leverage the scheduler cache’s existing resource accounting logic.

If the pod is already assigned to the node being evaluated, the scheduler’s cache has already factored the pod’s resource footprint into the node’s aggregated requested resources (nodeInfo.Requested), so the plugin uses nodeInfo.Requested directly in most cases.

However, in the event that the pod taken from the queue has resource requests diverging from the pod taken from the cache (which can happen due to some race conditions), the plugin will look up the pod in NodeInfo and adjust the node’s requested resources by adding the difference, assuming the pod’s resources to be the maximum of the two.

Handling Node Evaluation Results

In most cases, this restricted search results in a resource fit error, naturally triggering the scheduler’s preemption pathway. In this case, the pod will be kept in the Unschedulable queue.

However, if the resize fits without needing preemption (e.g., if cluster topology shifts or other workloads exit, freeing up capacity on the host), the NodeResourcesFit plugin returns a status of UnschedulableAndUnresolvable (with a message like "pod resize fits, waiting for Kubelet actuation"). Because the node status is UnschedulableAndUnresolvable rather than Unschedulable, the DefaultPreemption plugin’s PostFilter phase skips victim selection on the node (as it only evaluates Unschedulable nodes). This ensures the scheduler skips the preemption phase, but keeps the deferred pod in the Unschedulable queue.

Because a parked pod’s capacity can be consumed by other workloads before the Kubelet actuates it, the scheduler registers NodeResourcesFit QHints to watch for pod upsize events on the same node. If another pod on the node scales up, the QHint moves the parked pod back to the activeQ for re-evaluation.

The necessity of keeping the pod in the Unschedulable queue and using QHints for requeuing is explained further in Reconsideration Race Conditions .

DefaultPreemption Plugin: Preemption Mechanism Adjustments in the PostFilter Phase

Only the DefaultPreemption plugin runs during the PostFilter phase for deferred resize pods; other PostFilter plugins will be modified to skip processing for deferred resize pods.

The modifications to the NodeName and NodeResourceFit plugins ensure two important components of the desired behavior:

  • Victim Selection Constraint: The search for preemption victims is isolated exclusively to the deferred pod’s active node. This is already done by the NodeName plugin in the PreFilter phase. However, because of workload-aware group scheduling, preempting a victim on the target node might still trigger the cascaded preemption of related pods on other nodes.

  • Resource Accounting Adjustment: During the preemption evaluation, the fit plugin is adjusted to specially handle the deferred pod’s resources, as described in NodeResourcesFit Plugin: Calculating Resource Fit in the Filter Phase .

Similar to the Filter phase, the preemption and reprieve logic runs only the resource-fit checks, skipping irrelevant constraints like affinity and topology spread constraints.

Summary of Scheduling Cycle Flow

Taking all the above into account, the logic for processing a Deferred resize is as follows:

  1. Identify Deferred Status: Confirm the pod has a Deferred resize and is already bound to a node. Enqueue it in the Scheduling queue.
  2. Evaluate Fit and Trigger Preemption: Perform node evaluation restricted to the current node.
    • The node evaluation logic should run only the logic for the resource-fit check, skipping filters that are relevant only to initial scheduling, such as affinity/anti-affinity rules and topology spread constraints.
    • The resource-fit logic is adjusted to specially handle the deferred pod’s resources, as described in NodeResourcesFit Plugin: Calculating Resource Fit in the Filter Phase .
    • If the resize fits on the node, move the pod into the Unschedulable queue and skip victim selection (see Reconsideration Race Conditions for more details).
  3. Trigger Preemption If a FitError occurs, initiate the Scheduler preemption logic.
  4. Calculate Victims: Identify suitable preemption victims, with the search restricted to the current node. The resource-fit logic is adjusted to specially handle the deferred pod’s resources, as described in NodeResourcesFit Plugin: Calculating Resource Fit in the Filter Phase , and irrelevant constraints such as affinity and topology spread are skipped.
  5. Reevaluation: When the victim pod is removed, the Deferred resize pod is triggered and moved from Unschedulable pods into the scheduling queue, resulting in reevaluation.

Scheduler Resource Reservation

After a preemption occurs, we want to ensure that the space is not taken up by another pod that has not yet been scheduled. However, due to the fact that the Scheduler assumes that the resources being used by a pod are equal to max(desired, allocated, actual), and that the new requested resizes are reflected in the pod’s spec, the scheduler naturally already reserves the space for the resizing pod. This means that the Scheduler does not need to take any special action to ensure the resources are reserved.

Kubelet-Scheduler Preemption Interaction

The Kubelet monitors pod removals, including evictions, and automatically retries a resize as soon as it detects that a victim pod has been cleared. The removal of the victim pod triggers a scheduling cycle, during which the deferred resize pod is reevaluated.

Reconsideration Race Conditions

It is possible that the Kubelet chooses to resize a different pod than the one that triggered the eviction. Consider the case where Pod A has a deferred resize and triggers the eviction of Pod B. In the meantime, a higher-priority resize of Pod C is requested. In this case, the kubelet correctly prioritizes the resize of Pod C over the resize of Pod A. In this case, we ensure that Pod A stays in the scheduling queue until its resize can complete successfully, including any additional evictions it may trigger.

We can model deferred resize retries analogous to the way standard scheduling already works today. For standard pods, if a pod cannot be scheduled:

  1. The new pod triggers preemption.
  2. The pod gets moved to the Unschedulable queue.
  3. When the preemption victim is finally removed, pods that were previously unschedulable but may now become schedulable are moved back to the active (or backoff) queue.

With deferred resize, we can follow a parallel path:

  1. The resize request triggers preemption.
  2. The pod gets moved to the Unschedulable queue.
  3. When the preemption victim is removed, the deferred resize pod is moved back to the active (or backoff) queue.

However, a notable race condition arises: the Kubelet may not finish resizing either Pod A or Pod C before the scheduler evaluates Pod A again. During this reevaluation, the scheduler observes that the resize now ‘fits’ on the node during its node fit checks.

To protect against this race, we park the resize back into the Unschedulable queue. The lifecycle then resolves via one of three eventualities:

  • Pod A is correctly resized: It loses its deferred condition. The scheduler observes this through watches and discards it from tracking.
  • Pod C is resized: (Even though the scheduler preempted for Pod A). There is now no longer enough room for Pod A’s resize. The scale-up event of Pod C triggers the NodeResourcesFit QHint, which moves Pod A back to the activeQ so it can trigger another preemption (as explained in Handling Node Evaluation Results ).
  • Both Pod A and Pod C get resized: Both lose deferred conditions and are removed from scheduling queue tracking entirely.

The following diagram illustrates the flow:

graph TD
    A[Pod A Enters Queue] --> B{Does Resize Fit?}
    B -- No --> C[Calculate Preemption Victims]
    C --> D[Evict Pod B]
    D --> E[Victim Pod B Removed]
    
    E --> F{Did Kubelet actuate Pod A?}
    F -- Yes --> G[Remove Pod A from Tracking]
    F -- No: Space consumed by competing Pod C --> H[Return Pod A to Unschedulable Queue]
    
    B -- Yes: Space cleared but Kubelet hasn't actuated --> I[Park Pod A in Unschedulable Queue]
    I -- Pod C upsized: QHint triggers --> A

See alternative considerations for the deferred resize lifecycle here .

Preemption Policies

The existing preemption policies for scheduling new pods now also apply to Deferred resizes.

This means that the default preemptionPolicy, PreemptLowerPriority, allows the resize of pods of that PriorityClass to preempt lower-priority pods. If preemptionPolicy is set to Never, the resize of pods in that PriorityClass is non-preempting.

We considered creating a separate, independent preemption policy for Deferred resizes, but introducing this divergence adds unnecessary complexity for both users and maintainers:

  • Users would now be required to manage two different preemption behaviors for one pod.
  • The behavior may end up hard to predict, where a pod can preempt upon starting, but not to grow.
  • We would have to define the matrix of interactions between two independent preemption fields, leading to ambiguous states (e.g. what happens if preemptionPolicy is set to Never but resizePreemptionPolicy is set to PreemptLowerPriority)?
  • Adds code complexity, including updates to the Pod API, validation logic, and new branches in the Scheduler.

If a pod is marked with preemptionPolicy: PreemptLowerPriority, the user has already communicated that this workload is more important than lower-priority tasks. Whether that importance is manifested during initial placement or vertical scale-up, the intent of the priority remains the same.

See alternative considerations for preemption policies here .

Pod Priority, Graceful Termination, and Pod Disruption Budget

Because we are reusing the existing Scheduling preemption logic, the behavior is aligned with existing preemption behavior for new pods. This means:

  • Pods are preempted according to their Pod Priority.
  • The graceful termination period of victim pods is honored.
  • PodDisruptionBudget is supported, but not guaranteed.

Node-level Preemption Policy for In-Place Pod Resize

To avoid unnecessary disruption on nodes that have their own autoscaling solutions (such as resizable nodes), cluster operators and controllers need a way to disable scheduler-triggered preemption specifically when it is caused by an in-place pod resize request on those nodes. This protects critical workloads by preventing the scheduler from preempting low-priority pods to make room for a higher-priority resize, allowing the system to prefer node upsizing over preemption-induced disruption.

We propose a new field on the Node object to configure this policy, wrapped within a high-level PodPreemptionPolicy struct in NodeSpec:

// NodeSpec contains the desired configuration of a node.
type NodeSpec struct {
    // PodPreemptionPolicy controls the node-level preemption behaviors for pods on this node.
    // +optional
    PodPreemptionPolicy *NodePodPreemptionPolicy `json:"podPreemptionPolicy,omitempty"`
}

// NodePodPreemptionPolicy defines the node-level policies governing preemption for pods on this node.
type NodePodPreemptionPolicy struct {
    // DisableResizePreemption lists the owners (e.g., autoscalers, operators, administrators) 
    // that have requested to disable scheduler and Kubelet preemption for in-place pod resize on this node.
    // If this list is non-empty, resize-induced preemption is disabled on this node.
    // +listType=set
    // +optional
    // +k8s:optional
    // +k8s:eachVal=+k8s:format=k8s-label-key
    // +k8s:maxItems=20
    // +k8s:listType=set
    DisableResizePreemption []string `json:"disableResizePreemption,omitempty"`
}

An example YAML configuration for the Node:

apiVersion: v1
kind: Node
metadata:
  name: resizable-node
spec:
  podPreemptionPolicy:
    disableResizePreemption:
      - autoscaling.k8s.io/cluster-autoscaler
      - autoscaling.k8s.io/vpa-updater

The strings under disableResizePreemption have the same validation and constraints as standard Kubernetes label keys (e.g., a DNS subdomain prefix and a name separated by a /), and the list can have maximum of 20 entries.

See alternative considerations for the node-level preemption policy API here .

Kubelet-Scheduler Policy Coordination and Status Propagation

While having the scheduler inspect the Node field directly is technically feasible, we choose to have the Kubelet manage this policy and propagate it to the scheduler via the Pod status. This approach offers several key design advantages:

  1. Cleaner Decoupling: Keeping the scheduler’s preemption path pod-centric avoids direct coupling between the scheduler and individual Node specification policies.
  2. Lifecycle Alignment: The Kubelet natively owns the resize evaluation and enactment lifecycle. By letting the Kubelet evaluate the node policy and update the Pod status, the policy remains tightly integrated with the pod’s resize lifecycle.
  3. Kubelet Preemption Consistency: The Kubelet itself performs internal preemption for critical workloads (e.g., node-critical pods) requesting a resize. Having the Kubelet manage the policy enables Kubelet-side preemption to seamlessly honor the same preemption controls.

Instead, the mechanism relies on cooperation between the Kubelet and the Scheduler via the Pod status, leveraging the PodResizePreemptionDisabled condition which is described in more detail in the PodResizePreemptionDisabled Pod Condition section.

The cooperation between the Kubelet and Scheduler is as follows:

  1. Configuration: A cluster operator or controller adds its identifier to the spec.preemptionPolicy.disablePodResizePreemption set on the Node object.
  2. Kubelet Inspection: The Kubelet watches its own Node object and caches this configuration. If the preemption policy changes, the Kubelet updates the PodResizePreemptionDisabled condition on already-Deferred pods to reflect the new policy.
  3. Pod Status Update: When the Kubelet evaluates an In-Place Pod Resize request and determines it must be Deferred due to insufficient capacity, it checks the Node’s spec.preemptionPolicy.disablePodResizePreemption field.
  4. Signaling the Scheduler: If the list is non-empty, the Kubelet updates the Pod’s status by setting the Deferred resize condition and setting the Kubelet-owned PodResizePreemptionDisabled condition to True (with Reason: PreemptionDisabled and an appropriate message).
  5. Scheduler Action: The Scheduler, in its UpdatePod event handler and scheduling queue processing, inspects the Pod’s conditions. If a Deferred pod has PodResizePreemptionDisabled set to True with Reason PreemptionDisabled, the Scheduler ensures that the pod is not in the scheduling queue. The pod remains Deferred without triggering cluster disruption. If a Deferred pod does not have PodResizePreemptionDisabled set to True with Reason PreemptionDisabled, the scheduler ensures that the pod is in the scheduling queue and attempts preemption.

Multiple Owner Support

In environments with multiple controllers (e.g., multiple autoscalers) managing the same node, conflicts can arise. To support this, the spec.preemptionPolicy.disablePodResizePreemption field is designed as a set-type list (+listType=set). The Kubelet honors the policy if the list contains any entries (representing at least one active owner requesting the policy).

Policy Scope: Exclusivity to Pod Resize

This policy is strictly limited to pod resize requests and does not apply to new pod scheduling. In a mixed cluster (containing both resizable and non-resizable nodes), applying a preemption-disabling policy to new pod scheduling on a resizable node would cause the scheduler to simply select victims on a non-resizable node instead, defeating the purpose of the policy. For in-place pod resize, however, preemption is already strictly confined to the target node where the pod is running.

Kubelet Preemption

The Kubelet contains internal preemption logic to ensure that critical pods can be admitted and run. With In-Place Pod Resize, this extends to Kubelet-side preemption when a critical pod requests a resize that exceeds available node capacity.

Kubelet-side preemption also honors the node-level preemption policy. If the Node’s spec.podPreemptionPolicy.disableResizePreemption list is non-empty, the Kubelet does not preempt existing pods on the node to accommodate the resize request. Instead, the Kubelet marks the pod’s resize as Deferred and sets the PodResizePreemptionDisabled condition to True with reason PreemptionDisabledByNodePolicy.

This ensures a consistent operational model for the node: no pods are preempted to satisfy any resize request, preserving the stability of all workloads on the node and forcing the system to rely on node autoscaling (upsizing) to resolve the resource deficit.

PodResizePreemptionDisabled Pod Condition

When the Node’s preemption policy prohibits preemption for resize requests (the spec.podPreemptionPolicy.disableResizePreemption list is non-empty), the Kubelet sets the PodResizePreemptionDisabled condition to True with reason PreemptionDisabledByNodePolicy upon deferring the resize. This communicates to the scheduler that the Deferred resize request should be blocked because preemption has been disabled on the node.

status:
  conditions:
  - type: PodResizePreemptionDisabled
    status: "True"
    reason: PreemptionDisabledByNodePolicy
    message: "Preemption for in-place pod resize is disabled on node 'node-1' by preemption policy."
    lastTransitionTime: "2026-02-23T15:23:13Z"

The Kubelet clears this condition when the node’s preemption policy is updated to allow preemption (clearing the list), or when the pod’s Deferred resize is resolved or cancelled.

Scheduler Action

The Scheduler ignores Deferred pods that have the PodResizePreemptionDisabled condition set to True.

The Scheduler’s UpdatePod handler watches for pod updates and reacts to the PodResizePreemptionDisabled condition as follows:

  • Condition Added: When the PodResizePreemptionDisabled condition is set, the Scheduler removes the pod from the scheduling queue.
  • Condition Cleared: When the PodResizePreemptionDisabled condition is cleared but the pod’s resize is still Deferred, the Scheduler adds the pod back to the active scheduling queue to attempt preemption.

Failures and Reconsideration of Deferred pods

To ensure efficiency and avoid unnecessary reconciliation loops, the Scheduler treats Deferred resizes that fail preemption (when preemption is enabled) as a standard FitError (Unschedulable). By classifying the resize failure this way, the existing Scheduler retry and backoff logic is triggered automatically, requiring no new dedicated retry infrastructure.

  1. Queue Placement: When a preemption attempt fails to find enough victims on the node to satisfy the resize, the pod is moved to the unschedulablePods pool within the scheduling queue.
  2. Triggering Re-entry: Because the pod is in the unschedulablePods pool, it is moved back to the activeQ (or backoffQ) automatically when cluster events occur that could potentially change the outcome of the preemption logic. These events include:
    • Pod Deletion: A pod on the same node is deleted or removed (freeing up capacity).
    • Pod Downsize: A pod on the same node has its resource requests decreased.
    • Node Update: The node’s Allocatable resources increase (e.g., via a Kubelet config update).
  3. Backoff Mechanism: Standard exponential backoff applies. This ensures that “stuck” resize requests do not degrade the scheduling throughput for the rest of the cluster.

By utilizing the existing movePodsToActiveOrBackoffQueue logic and treating the resize deficit as a native scheduling constraint, we ensure that Deferred resizes do not consume excessive CPU cycles unless the cluster state has changed in a way that makes success plausible.

To prevent ’thundering herd’ issues where unrelated node activity causes resizes to be reevaluated unnecessarily, a pod is only moved back to the activeQ when the pod deletion, pod downsize, or node update events occur on the same node as the pod being resized. This is implemented by making the QueueingHint (Qhint) functions of the NodeResourcesFit plugin node-aware; if the pod is already assigned to a node (indicating a resize request), the Qhints evaluate to QueueSkip for any event occurring on a different node.

Failure Handler Adjustments for Deferred Pods

When a deferred resize scheduling cycle fails (i.e., there is a fit error and preemption fails or is disabled), the scheduler executes its failure handler (handleSchedulingFailure). Because the deferred pod is already scheduled and running on a node, several steps in the standard failure handler must be modified or skipped:

  • Bypassing the NodeName Re-queue Abort: For a standard pod, if the pod already has Spec.NodeName populated in the informer cache, the failure handler assumes it has been successfully bound and aborts re-queuing it. For Deferred pods, this abort check is skipped so that the pod is successfully returned to the Unschedulable queue to await future retry attempts.
  • Skipping PodScheduled Condition Updates: The standard failure handler sets the pod’s PodScheduled status condition to False (with reason Unschedulable). For a Deferred pod, updating this condition is skipped. For beta, we will implement emitting events to surface the results of the scheduling cycle for observability.
  • No Nominated Node Name (NNN) Assignment: Standard scheduling failures can record a NominatedNodeName on the pod status to reserve space on a target candidate node. For a Deferred pod, no nominated node is set or updated. The pod is already bound to its host, and its resize can only be evaluated and satisfied on that specific host; nominating another node is invalid.

Scope of Interaction with Workload-Aware Preemption

With workload-aware preemption, there are two scenarios to consider:

  1. The selected victim pod is part of a workload: The scheduler preempts the entire workload. Since the preempted workload’s priority is strictly lower than that of the resizing pod, this behaves as designed.

  2. The resizing pod itself is part of a workload:

    • High-Level Direction: When a resizing pod belongs to a workload group (e.g., a PodGroup scheduled atomically), resize-induced preemption must evaluate the impact on the entire workload group. The scheduler must verify if the resize violates any workload-wide invariants (such as minimum member availability or atomic group scheduling rules).
    • Queueing Behavior for Group Members: A running pod that requests a resize must be queued and retried individually. If the pod belongs to a PodGroup, the scheduling queue instead treats the deferred pod as an independent individual pod. In Alpha, this will be implemented by having isPodGroupMember return false for deferred resize pods, allowing them to follow the standard individual queueing and backoff pathways.
    • Alpha Scope: The resizing pod is evaluated individually for preemption victim selection on its assigned node. The scheduler does not proactively trigger group-wide rescheduling or preemption of other members of the workload group.
    • Beta Graduation: Co-existence mechanics, including group-wide coordinated preemption (e.g., preempting other members of the same workload to balance resource usage or preventing preemption if the workload’s group-wide health is already degraded), will be fully designed and finalized prior to Beta.

Test Plan

[x] I/we understand the owners of the involved components may require updates to existing tests to make this code solid enough prior to committing the changes necessary to implement this enhancement.

Prerequisite testing updates
Unit tests
  • k8s.io/kubernetes/pkg/apis/scheduling/v1alpha1: 2026-02-23 - 62.7%
  • k8s.io/kubernetes/pkg/apis/scheduling/validation: 2026-02-23 - 97.8%
  • k8s.io/kubernetes/pkg/scheduler: 2026-02-23 - 81.7%
Integration tests

In alpha, we will create integration test(s) to ensure basic functionalities:

  • A lower-priority pod can be preempted by a higher-priority Deferred resize.
  • Several lower-priority pods can be preempted by a higher-priority Deferred resize.
  • Non-preempting pods according to the preemptionPolicy do not trigger preemption.

For beta, we will add additional, more comprehensive integration test. The list here will be finalized here prior to beta graduation.

e2e tests

We will create a single e2e test to ensure that a lower-priority pod can be preempted by a higher-priority Deferred resize, and that the Kubelet successfully actuates the Deferred resize after the preemption completes.

Graduation Criteria

Alpha

  • Feature and API implemented behind a feature flag
  • Initial unit, integration, and e2e tests completed and enabled

Beta

  • Gather feedback from alpha
  • Metrics and events are defined and implemented
  • Additional integration tests are defined, implemented and linked in KEP
  • Interaction with workload aware scheduling is clarified
  • Address scenarios where a preemption victim’s grace period exceeds the time window allocated for executing the resize operation (such as VPA’s fallback time limit).
  • Evaluate using nominated node name to prevent double preemption (specifically for the shifting preemption victims scenario); see alternatives in Tracking Preemption Nominations to Avoid Double Preemption .
  • Evaluate whether we need to solve the additional preemption risk (see Risk of Additional Preemption ).
  • If KEP-5517 Alpha2 is implemented, ensure that the Deferred resize scheduling cycle includes resources from pod.status.nodeAllocatableResourceClaimStatuses.

GA

  • Allowing time for feedback
  • All issues and gaps identified as feedback during beta are resolved
  • Additional GA requirements TBD at Beta release

Upgrade / Downgrade Strategy

Standard procedures for features introducing new API fields should be used:

  • On upgrade, kube-apiservers should be upgraded first before the Kubelet can write, and kube-scheduler can read, the new pod condition PodResizePreemptionDisabled and the new node-level PodPreemptionPolicy field.
  • On downgrade, kube-schedulers and Kubelets should be downgraded first (to stop using and writing the new fields) before kube-apiservers are downgraded; note that downgrade of kube-apiserver(s) and/or disabling the new API fields will not clear their contents for objects already stored in the storage (etcd).

Version Skew Strategy

N-3 kubelet already marks resizes without enough capacity as Deferred and retries when room is made.

While an n-3 Kubelet correctly defers resizes, it does not recognize the new spec.podPreemptionPolicy Node field or propagate the PodResizePreemptionDisabled condition.

To bridge this version skew gap and enforce the operator’s policy during upgrade windows, the Scheduler will temporarily check the Node’s spec.podPreemptionPolicy field directly when evaluating Deferred pod resizes. Once Kubelets are upgraded to a version supporting this feature, they will assume sole ownership of policy propagation via the pod condition, and this temporary Scheduler fallback will be removed in a future release.

Production Readiness Review Questionnaire

Feature Enablement and Rollback

How can this feature be enabled / disabled in a live cluster?
  • Feature gate (also fill in values in kep.yaml)
    • Feature gate name: SchedulerPreemptionForPodResize
    • Components depending on the feature gate: kube-apiserver, kube-scheduler, kubelet
Does enabling the feature change any default behavior?

If users already have PriorityClasses defined in their cluster, and are already using In-Place Pod Resize, Deferred resizes will now trigger preemption of lower-priority pods.

Can the feature be disabled once it has been enabled (i.e. can we roll back the enablement)?

Yes, the feature can be rolled back by disabling the feature gate.

What happens if we reenable the feature if it was previously rolled back?

The feature will start working again; Deferred resizes will trigger preemption of lower-priority pods.

Are there any tests for feature enablement/disablement?

The scheduler logic will be covered by regular feature tests.

For the newly introduced API, dedicated enablement/disablement tests at the kube-apiserver registry layer will be added in alpha release.

Rollout, Upgrade and Rollback Planning

How can a rollout or rollback fail? Can it impact already running workloads?
What specific metrics should inform a rollback?
Were upgrade and rollback tested? Was the upgrade->downgrade->upgrade path tested?
Is the rollout accompanied by any deprecations and/or removals of features, APIs, fields of API types, flags, etc.?

Monitoring Requirements

How can an operator determine if the feature is in use by workloads?
How can someone using this feature know that it is working for their instance?
  • Events
    • Event Reason:
  • API .status
    • Condition name:
    • Other field:
  • Other (treat as last resort)
    • Details:
What are the reasonable SLOs (Service Level Objectives) for the enhancement?
What are the SLIs (Service Level Indicators) an operator can use to determine the health of the service?
  • Metrics
    • Metric name:
    • [Optional] Aggregation method:
    • Components exposing the metric:
  • Other (treat as last resort)
    • Details:
Are there any missing metrics that would be useful to have to improve observability of this feature?

Dependencies

Does this feature depend on any specific services running in the cluster?

Scalability

Will enabling / using this feature result in any new API calls?

For most of the logic, we are not introducing any new API calls. The Scheduler is already watching for pod updates. The node and pod states are cached in the Scheduler’s memory, and is updated when the Informer receives an update. The kubelet will update the pod status when the resize is Deferred due to insufficient capacity; this is not a new API call.

The kubelet is already watching for node updates. However, the kubelet will now make new API calls to update the pod statuses of already-deferred pods in response to changes in the node-level preemption policy.

Will enabling / using this feature result in introducing new API types?

No.

Will enabling / using this feature result in any new calls to the cloud provider?

No.

Will enabling / using this feature result in increasing size or count of the existing API objects?

The new condition in the pod status will increase the size of pod objects, only when the condition is present.

Will enabling / using this feature result in increasing time taken by any operations covered by existing SLIs/SLOs?

There may be some performance impact for scheduling of the pods, as the scheduler now has to additionally process Deferred pods.

Will enabling / using this feature result in non-negligible increase of resource usage (CPU, RAM, disk, IO, …) in any components?

No.

Can enabling / using this feature result in resource exhaustion of some node resources (PIDs, sockets, inodes, etc.)?

No.

Troubleshooting

How does this feature react if the API server and/or etcd is unavailable?
What are other known failure modes?
What steps should be taken if SLOs are not being met to determine the problem?

Implementation History

2026-02-23: KEP Created for alpha release

Drawbacks

Alternatives

Separate scheduling queue

We considered keeping Deferred resizes in a separate scheduling queue from new pods. However, the interaction and prioritization between these two queues was difficult to reason about. It is simpler and more intuitive to consider Deferred resizes as warranting largely the same behavior as new pods with regards to preemption, so keeping them together in a single queue made more sense.

Implementing prioritized resizes logic

The Kubelet has a priority order that it attempts to retry Deferred resizes, as defined here .

We considered integrating a similar prioritization logic for the Deferred resizes in the scheduling queue. However, this adds a lot of complexity for questionable benefit. Scheduler preemption today considers only Pod Priority, and unless there is a strong use case to do something more complex, the preemption behavior for Deferred resizes should stay aligned with that of new pods.

Preemption Policies

We considered creating a separate, independent preemption policy for Deferred resizes, but introducing this divergence adds unnecessary complexity for both users and maintainers:

  • Users would now be required to manage two different preemption behaviors for one pod.
  • The behavior may end up hard to predict, where a pod can preempt upon starting, but not to grow.
  • We would have to define the matrix of interactions between two independent preemption fields, leading to ambiguous states (e.g. what happens if preemptionPolicy is set to Never but resizePreemptionPolicy is set to PreemptLowerPriority)?
  • Adds code complexity, including updates to the Pod API, validation logic, and new branches in the Scheduler.

If a pod is marked with preemptionPolicy: PreemptLowerPriority, the user has already communicated that this workload is more important than lower-priority tasks. Whether that importance is manifested during initial placement or vertical scale-up, the intent of the priority remains the same.

Conflicting Sources of Truth for the Pod

To address the complexity of the pod existing in both the cache and the scheduling queue simultaneously, we considered avoiding duplicate state by using a dynamic reference interface.

Instead of storing a direct snapshot of the pod in the scheduling queue, we considered changing the pod representation inside the queue to a PodGetter interface:

type PodGetter interface {
    GetPod() *v1.Pod
}

Under this alternative:

  • For standard pods, GetPod() would simply return the stored pod snapshot.
  • For deferred resize pods, GetPod() would dynamically query the pod directly from the scheduler’s shared cache, ensuring a single source of truth.

While this abstraction avoids the need to synchronize pod updates across multiple places, we rejected it because:

  • Bypassing critical scheduling code paths: Storing a dynamic reference might skip essential queue event handlers. For example, PriorityQueue.Update would not be called correctly upon pod updates. This could prevent a pod’s preemption from being retried when a relevant event (such as a status or request change) is updated.
  • Increased codebase management overhead: Circumventing the standard queue update flow would make the scheduling queue’s state harder to reason about and manage.
  • Complex refactoring footprint: It introduces a significant refactoring footprint across core scheduler structures and scheduling plugins that expect immutable pod snapshots throughout their scheduling cycle.

Instead, we chose to explicitly keep the pod object in both places (the cache and the queue) while ensuring we audit and eliminate any downstream logic that assumes the pod exists in only one place.

Kubelet-Scheduler Preemption Interaction

We considered introducing a mechanism to communicate to the Kubelet exactly which resize event initiated a preemption cascade. This would eliminate opportunistic race conditions between the Kubelet and the scheduler. However, this adds a lot of complexity for arguably worse behavior.

For example, if a higher-priority resize is requested while room is being made by evicting a lower-priority pod, we want to honor the priority. We want the higher-priority pod to get resized, and the other deferred pod to be put back into the scheduling queue. By avoiding explicit signaling, we let Kubernetes naturally honor pod priority; the Kubelet can allocate capacity to the highest-priority workload, and the displaced pod safely returns to the scheduling queue.

Tracking Preemption Nominations to Avoid Double Preemption

If a Deferred resize triggers preemption, there is a risk of “double preemption” (where the scheduler terminates a different set of victims for the same resize request) in two scenarios:

  1. Scheduler Restart: The scheduler restarts mid-preemption, clearing its in-memory preemption state, and selects a new victim pod upon recovery.
  2. Long Grace Periods: A victim has a long graceful termination period. While it is terminating, other node activity (such as pod deletion) triggers the scheduler to re-evaluate the Deferred pod. If the cluster state has changed or selection is non-deterministic, the preemption logic might select a new victim instead of recognizing the existing termination process.

To mitigate this, the scheduler needs a way to track which pod is preempting what via resource nominations so it does not retry preemption while the initial victim is still terminating. We considered four potential options to address this, and will finalize the approach during the Beta phase:

Option 1: Accept Double Preemption (Alpha Decision)

  • Description: Treat double preemption as an acceptable edge-case behavior, relying on the API server’s delete idempotency to avoid duplicate evictions of the same victim.
  • Pros: Zero implementation complexity; keeps the scheduler stateless.
  • Cons: Can lead to redundant preemption of additional low-priority workloads in rare scenarios, generating unnecessary load and workload disruption.

Option 2: Internal Nomination Tracking (In-Memory struct)

  • Description: Maintain an internal, scheduler-private map/struct that tracks active preemption nominations for Deferred resizes without persisting or exposing this data outside the scheduler.
  • Pros: Restricts the state strictly within the scheduler; does not affect other API clients.
  • Cons: Does not survive scheduler restarts, meaning it only resolves the long grace period scenario and leaves the scheduler restart scenario unmitigated.

Option 3: Reuse NominatedNodeName (NNN)

  • Description: Reuse the existing status.nominatedNodeName field on the Pod object to nominate the node where preemption is occurring, allowing the scheduler to recognize existing nominations across restarts and grace periods.
  • Pros: Resolves both scheduler restart and long grace period scenarios cleanly by leveraging standard preemption structures.
  • Cons: Semantically changes the meaning of nominatedNodeName (which is traditionally used only for unscheduled/unplaced pods, whereas the Deferred resizing pod is already bound to a node). This could break downstream assumptions in external components that monitor this field.

Option 4: Check Pod spec.nodeName in DefaultPreemption

  • Description: Modify the preemption logic in DefaultPreemption to treat spec.nodeName as the nomination node for already-bound pods with deferred resize requests.
  • Pros: Avoids mutating status.nominatedNodeName (preventing breaking downstream clients) and naturally survives scheduler restarts since spec.nodeName and the Deferred resize state are persisted.
  • Cons: Requires modifying DefaultPreemption to check both nominatedNodeName (for unscheduled pods) and spec.nodeName (for bound resizing pods) when evaluating active nominations.

We plan to evaluate the implications of Option 3 and Option 4 on the Kubernetes ecosystem and decide on the final design for Beta.

Node-Level Preemption Policy API Options

We identified several API options for implementing the node-level preemption policy before selecting the proposed first-class Node Field.

1. Node Annotations (Rejected)

  • Description: Setting the policy via a standard or prefixed annotation, such as scheduler.policy/disable-pod-resize-preemption: "true".
  • Why Rejected: Core Kubernetes components (such as the Scheduler and Kubelet) should generally not depend on unstructured annotations to drive critical runtime decisions. Annotations are meant for metadata and not for policy enforcement.

2. Scheduler Honors Node Field Directly (Rejected)

  • Description: The scheduler would directly watch Node objects, check for the preemption policy in the Node spec, and skip preemption directly.
  • Why Rejected: While this is technically feasible, it is better for the Kubelet to manage node-level policies and communicate them to the scheduler via the Pod status (e.g., setting the PodResizePreemptionDisabled condition to True with reason PreemptionDisabled on the Pod). This keeps the scheduler’s preemption path pod-centric, avoids direct coupling between the scheduler and individual Node spec policies, and ensures that Kubelet-side internal preemption can easily honor the same policy.

3. Node Labels (Rejected)

  • Description: Setting the policy via a node label, e.g., kubectl label node <node-name> scheduler.policy/disable-pod-resize-preemption="true".
  • Why Rejected: Labels are intended for identifying and grouping objects (e.g., for node selectors or affinity). They are not meant for carrying fine-grained configuration or policy intent. They lack structured representation and do not support multiple owners cleanly.

4. Separate API Object (Rejected)

  • Description: Defining a new API object (e.g., a NodePreemptionPolicy CRD) to manage preemption policies outside the core Node object, linked via a node selector.
  • Why Rejected: This approach is too heavyweight for a simple boolean toggle. It introduces the overhead of managing a new API type, implementing new controllers, and increasing memory usage in the control plane to cache these objects. The operational complexity outweighs the benefits.

5. Node Condition (Rejected)

  • Description: Representing the policy status using a Node Condition, e.g., type: PodResizePreemptionDisabled, status: "True".
  • Why Rejected: Node Conditions are designed to report the current observable state of a node (e.g., Ready, DiskPressure) as determined by the Kubelet or other controllers. They are not intended to be used by operators to configure desired behavior. Furthermore, conditions do not easily support the “multiple owner” scenario (e.g., multiple autoscalers) where multiple entities need to independently express their intent without overwriting each other.