# Kubernetes v1.35.0

2025-12-17 · [Original release notes](https://github.com/kubernetes/kubernetes/releases/tag/v1.35.0) · [Full summary](https://ratatosk.io/en/releases/kubernetes/v1.35.0)

Security 5 · Breaking changes 60 · Deprecations 2 · Important fixes 54 · Enhancements 51 · Dependency updates 99 · Other changes 74

## Check before upgrading

### Prerequisites

- **Partitionable Devices alpha compatibility change**: ResourceSlices using the feature must be removed before upgrading or downgrading between v1.34 and v1.35.
  > - Updated the Partitionable Devices feature to support referencing counter sets across ResourceSlices within the same resource pool. Devices from incomplete pools were no longer considered for allocation. This change introduced backwards-incompatible updates to the alpha feature, requiring any ResourceSlices using it to be removed before upgrading or downgrading between v1.34 and v1.35. ([#134189](https://github.com/kubernetes/kubernetes/pull/134189), [@mortent](https://github.com/mortent)) [SIG API Machinery, Node, Scheduling and Testing]

  https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

## Security (5)

### Kube-apiserver blocks request-header spoofing with shared client CAs

- Affects: Kube-apiserver configurations with overlapping certificates in `--requestheader-client-ca-file` and `--client-ca-file`
- Additional action: When the CA files overlap, `--requestheader-allowed-names` must be specified.

Previously, a client certificate issued by a shared CA could spoof identities through request-header authentication. The API server now aborts startup if the CA bundles overlap and `--requestheader-allowed-names` is unset.

> - Kube-apiserver: Made sure that when `--requestheader-client-ca-file` and `--client-ca-file` contain overlapping certificates, `--requestheader-allowed-names` must be specified so that regular client certificates cannot set authenticating proxy headers for arbitrary users. ([#131411](https://github.com/kubernetes/kubernetes/pull/131411), [@ballista01](https://github.com/ballista01)) [SIG API Machinery, Auth and Security]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### CSI service-account tokens can use the secrets field

- Affects: CSI drivers receiving service-account tokens, with `spec.serviceAccountTokenInSecrets: true` and the `CSIServiceAccountTokenSecrets` feature gate enabled

Previously, CSI service-account tokens were passed through volume context, exposing them in logs and other outputs. With the gate enabled and the driver opted in, kubelet passes tokens through secrets instead; drivers that do not opt in retain the existing behavior.

> - CSI drivers can now opt in to receive service account tokens via the secrets field instead of volume context by setting `spec.serviceAccountTokenInSecrets: true` in the CSIDriver object. This prevents tokens from being exposed in logs and other outputs. The feature is gated by the `CSIServiceAccountTokenSecrets` feature gate (beta in `v1.35`). ([#134826](https://github.com/kubernetes/kubernetes/pull/134826), [@aramase](https://github.com/aramase)) [SIG API Machinery, Auth, Storage and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Pod exec, attach, and port-forward requests require `create` permission

- Affects: custom ClusterRoles or Roles granting `pods/exec`, `pods/attach`, or `pods/portforward`
- Additional action: `AuthorizePodWebsocketUpgradeCreatePermission` feature gate, enabled by default

Previously, WebSocket requests for these subresources needed only `get`, while SPDY requests already needed `create`. After the upgrade, WebSocket requests also need `create`.

> - Kube-apiserver: Made the subresources `pods/exec`, `pods/attach`, and `pods/portforward` require `create` permission for both SPDY and Websocket API requests. Previously, SPDY requests required `create` permission, but Websocket requests only required `get` permission. This change is gated by the `AuthorizePodWebsocketUpgradeCreatePermission` feature-gate, which is enabled by default.
>   Before upgrading to 1.35, ensure any custom ClusterRoles and Roles intended to grant `pods/exec`, `pods/attach`, or `pods/portforward` permission include the `create` verb. ([#134577](https://github.com/kubernetes/kubernetes/pull/134577), [@seans3](https://github.com/seans3)) [SIG API Machinery, Auth, Node and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Higher APF seat usage for legacy watch calls

- Affects: legacy watch calls with `ResourceVersion` = 0 or unset that generate init-events

These calls may be throttled more. The higher seat usage accounting is intended to protect the API server from CPU overload.

> - Made legacy watch calls (`ResourceVersion` = 0 or unset) that generate init-events weigh higher in `API Priority and Fairness (APF)` seat usage. Properly accounting for their cost protects the API server from CPU overload. Users might see increased throttling of such calls as a result. ([#134601](https://github.com/kubernetes/kubernetes/pull/134601), [@shyamjvs](https://github.com/shyamjvs))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Untolerated taint details removed from Pod status messages

- Affects: Pod scheduling with taints

When a Pod failed scheduling because it did not tolerate a taint, its status message included the taint’s specific key and value. The message now says “node(s) had untolerated taint(s)”; the specific taint information appears in logs.

> - `kube-scheduler`: Pod statuses no longer include specific taint keys or values when scheduling fails due to untolerated taints. ([#134740](https://github.com/kubernetes/kubernetes/pull/134740), [@hoskeri](https://github.com/hoskeri))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

## Breaking changes (60)

### Pod resize validation now rejects non-CPU and non-memory resources

- Affects: pod resize requests

Previously, a resize could add a resource other than CPU or memory when the container had no resources. Now, that request fails with a validation error.

> - Pod resize now only allows CPU and memory resources; other resource types are forbidden. ([#135084](https://github.com/kubernetes/kubernetes/pull/135084), [@tallclair](https://github.com/tallclair)) [SIG Apps, Node and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### HPA configurable tolerance promoted to beta with the `HPAConfigurableTolerance` gate enabled by default

- Affects: HorizontalPodAutoscalers with configurable tolerance settings (Inferred)

The feature was alpha before this release. HPA scaling rules can now use tolerance without separately enabling the gate.

> - Promoted HPA configurable tolerance to beta. The `HPAConfigurableTolerance` feature gate has now been enabled by default. ([#133128](https://github.com/kubernetes/kubernetes/pull/133128), [@jm-franc](https://github.com/jm-franc)) [SIG API Machinery and Autoscaling]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Separate alpha gate for `DeviceTaintRules` support

- Affects: DRA device taints and tolerations

With the new gate at its default false setting, `DeviceTaintRules` support is disabled. `DRADeviceTaints` can remain enabled, so tainting through `ResourceSlices` continues to work.

> - The DRA device taints and toleration feature received a separate feature gate, `DRADeviceTaintRules`, which controlled support for `DeviceTaintRules`. This allowed disabling it while keeping `DRADeviceTaints` enabled so that tainting via `ResourceSlices` continued to work. ([#135068](https://github.com/kubernetes/kubernetes/pull/135068), [@pohly](https://github.com/pohly)) [SIG API Machinery, Apps, Auth, Node, Scheduling and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Pod status and startup probes stay unchanged on kubelet restart

- Affects: Pods managed by kubelet (Inferred)

Kubelet restarts could cause Pod status changes.

> - Added the `ChangeContainerStatusOnKubeletRestart` feature gate, which defaults to disabled. When the feature gate is disabled, `kubelet` does not change the Pod status upon restart, and Pods do not re-run startup probes after the `kubelet` restarts. ([#134746](https://github.com/kubernetes/kubernetes/pull/134746), [@HirazawaUi](https://github.com/HirazawaUi)) [SIG Node and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### API server SLI metric excludes `dryRun` request latencies

- Affects: API-server metric consumers using `apiserver_request_sli_duration_seconds` with `dryRun` traffic (Inferred)

Other requests continue to be recorded in the metric.

> - Metrics: Excluded `dryRun` requests from `apiserver_request_sli_duration_seconds`. ([#131092](https://github.com/kubernetes/kubernetes/pull/131092), [@aldudko](https://github.com/aldudko)) [SIG API Machinery and Instrumentation]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `HostnameOverride` promoted to beta and enabled by default

- Affects: workloads using `HostnameOverride` (Inferred)

In v1.34, `HostnameOverride` was alpha and disabled by default.

> - Promoted the `HostnameOverride` feature gate to beta and enabled it by default. ([#134729](https://github.com/kubernetes/kubernetes/pull/134729), [@HirazawaUi](https://github.com/HirazawaUi)) [SIG Network and Node]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `kubectl describe pods` adds field paths to event messages

- Affects: pod events with an involved object's `fieldPath`

Previously, these messages omitted the path; it now appears before the event text.

> - Updated `kubectl describe pods` to include the involved object’s `fieldPath` (e.g., container name) in event messages, providing better context for debugging multi-container Pods. Note: This changes the previous message format for events that include a `fieldPath`. ([#133627](https://github.com/kubernetes/kubernetes/pull/133627), [@itzPranshul](https://github.com/itzPranshul))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Deprecated metrics hidden at policy deadlines

- Affects: deprecated metrics that have reached their policy hiding point (Inferred)

Alpha and Internal metrics are hidden in the minor release they are deprecated; Beta metrics one minor release after their `DeprecatedVersion`; and Stable metrics three minor releases after their `DeprecatedVersion`.

> - Deprecated metrics will be hidden as per the metrics deprecation policy. https://kubernetes.io/docs/reference/using-api/deprecation-policy/#deprecating-a-metric . ([#133436](https://github.com/kubernetes/kubernetes/pull/133436), [@richabanker](https://github.com/richabanker)) [SIG Architecture, Instrumentation and Network]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `SchedulerAsyncAPICalls` disabled by default

- Affects: `SchedulerAsyncAPICalls` enabled in `kube-scheduler`

Under high `kube-apiserver` load, asynchronous preemption could degrade `kube-scheduler` performance; in v1.34, the feature gate was enabled by default.

> - Disabled the `SchedulerAsyncAPICalls` feature gate to mitigate a bug where its interaction with asynchronous preemption could degrade `kube-scheduler` performance, especially under high `kube-apiserver` load. ([#134400](https://github.com/kubernetes/kubernetes/pull/134400), [@macsko](https://github.com/macsko))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Unused `DeviceBindingConditions` fields dropped when `DRADeviceBindingConditions` is disabled

- Affects: ResourceClaims with unused `DeviceBindingConditions` fields and `DRADeviceBindingConditions` disabled (Inferred)

> - Dropped `DeviceBindingConditions` fields when the `DRADeviceBindingConditions` feature gate is not enabled and not in use. ([#134964](https://github.com/kubernetes/kubernetes/pull/134964), [@sunya-ch](https://github.com/sunya-ch))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Object-size statistics skipped for resources outside the watch cache

- Affects: resources not served from the watch cache, typically `Events`

In v1.34, key listing for object-size statistics could be expensive and hurt `Events` POST latency. For events, `apiserver_resource_size_estimate_bytes` is unavailable, so APF estimates LIST request cost using a 1.5MB fallback and assigns requests `maxSeats`.

Evidence: “* No apiserver_resource_size_estimate_bytes metric for events. * APF overestimating LIST request cost to events. Fallback assumes object size of 1.5MB, meaning LIST events will always get maxSeats” https://github.com/kubernetes/kubernetes/pull/133873

> - Kube-apiserver: Fixed a possible `v1.34` performance regression calculating object size statistics for resources not served from the watch cache, typically only `Events`. ([#133873](https://github.com/kubernetes/kubernetes/pull/133873), [@serathius](https://github.com/serathius)) [SIG API Machinery and Etcd]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Kubelet pod hash generation switched to FNV-1a for empty UIDs

- Affects: pods with an empty UID that receive kubelet defaults

Previously, `applyDefaults` generated the hash for pods without a UID using MD5.

Evidence: “func applyDefaults(logger klog.Logger, pod *api.Pod, source string, isFile bool, nodeName types.NodeName) error {” https://github.com/kubernetes/kubernetes/pull/133511

> - Eliminated and prevented future use of the `md5` algorithm in favor of more appropriate hashing algorithms. ([#133511](https://github.com/kubernetes/kubernetes/pull/133511), [@BenTheElder](https://github.com/BenTheElder)) [SIG Apps, Architecture, CLI, Cluster Lifecycle, Network, Node, Security, Storage and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Container names removed from created and started event messages

- Affects: Kubelet events for created and started containers

Previously, these messages included the container name; they now read “Container created” or “Container started.”

> - Removed container name from messages for container created and started events. ([#134043](https://github.com/kubernetes/kubernetes/pull/134043), [@HirazawaUi](https://github.com/HirazawaUi))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `KUBECTL_OPENAPIV3_PATCH` no longer controls OpenAPI v3 patching

- Affects: kubectl setups with `KUBECTL_OPENAPIV3_PATCH` set (Inferred)

Previously, the variable could disable OpenAPI v3 patching. kubectl now attempts to use the OpenAPI v3 client without checking it.

> - Removed the `KUBECTL_OPENAPIV3_PATCH` environment variable, as aggregated discovery has been stable since `v1.30`. ([#134130](https://github.com/kubernetes/kubernetes/pull/134130), [@ardaguclu](https://github.com/ardaguclu))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### User-namespace PSA relaxation is now unconditional

- Affects: clusters applying PSA to user-namespace pods or specifying `UserNamespacesPodSecurityStandards` (Inferred)
- Action required: kubelet v1.31 or later

Previously, the gate prevented PSA policies from being relaxed for user-namespace pods before kubelet support was assured.

Evidence: “Now, we can unconditionally relax PSA policy if a pod is in a user namespace.” https://github.com/kubernetes/kubernetes/pull/132157

> - Removed the `UserNamespacesPodSecurityStandards` feature gate. The minimum supported Kubernetes version for `kubelet` is now `v1.31`, so the gate is no longer needed. ([#132157](https://github.com/kubernetes/kubernetes/pull/132157), [@haircommander](https://github.com/haircommander)) [SIG Auth, Node and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Referenced-token and secret counts removed from ServiceAccount output

- Affects: ServiceAccounts viewed with `kubectl get` or `kubectl describe` (Inferred)

The human-readable output previously included counts for referenced tokens and secrets; those counts no longer appear.

> - Updated `kubectl get` and `kubectl describe` human-readable output to no longer show counts for referenced tokens and secrets. ([#117160](https://github.com/kubernetes/kubernetes/pull/117160), [@liggitt](https://github.com/liggitt)) [SIG CLI and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### [No downgrade] DRA `resource.k8s.io` storage defaults to `v1`

- Affects: clusters using the DRA `resource.k8s.io` API and its stored resources

Downgrading to 1.33 is not supported.

> - DRA: the resource.k8s.io API now uses the v1 API version (introduced in 1.34) as default storage version. Downgrading to 1.33 is not supported. ([#133876](https://github.com/kubernetes/kubernetes/pull/133876), [@kei01234kei](https://github.com/kei01234kei)) [SIG API Machinery, Etcd and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `kubeadm` preflight errors on cgroups v1 with `v1.35` or newer kubelets

- Affects: hosts using cgroups v1 with `kubeadm` and kubelet v1.35 or newer
- Action required: Ignoring the `SystemVerification` preflight error and setting `failCgroupV1: false` in the `kube-system/kubelet-config` ConfigMap before upgrading
- Removed in: not stated

The validator previously warned when it detected cgroups v1. From v1.35, `failCgroupV1` is set to true, so nodes no longer start on cgroups v1 by default.

>  - ACTION REQUIRED:
>   vendor: Updated `k8s.io/system-validators` to `v1.12.1`. The cgroups validator now throws an error instead of a warning if cgroups v1 is detected on the host and the provided KubeletVersion is `v1.35` or newer.
>   kubeadm: Started using `k8s.io/system-validators` `v1.12.1` in `kubeadm` `v1.35`. During `kubeadm init`, `kubeadm join`, and `kubeadm upgrade`, the SystemVerification preflight check throws an error if cgroups v1 is detected and the detected `kubelet` version is `v1.35` or newer. For older versions of `kubelet`, a preflight warning is displayed.
>   To allow cgroups v1 with `kubeadm` and `kubelet` version `v1.35` or newer, you must:
>   - Ignore the error from the SystemVerification preflight check by `kubeadm`.
>   - Edit the `kube-system/kubelet-config` ConfigMap and add the `failCgroupV1: false` field before upgrading. ([#134744](https://github.com/kubernetes/kubernetes/pull/134744), [@neolit123](https://github.com/neolit123)) [SIG Cluster Lifecycle and Node]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### cgroup v1 nodes blocked from starting by default

- Affects: nodes using cgroup v1, especially with kubelet v1.35 or newer
- Action required: For kubeadm and kubelet v1.35 or newer, allowing cgroup v1 requires ignoring the kubeadm SystemVerification preflight error and adding `failCgroupV1: false` to the `kube-system/kubelet-config` ConfigMap before upgrading.
- Removed in: not stated

> - ACTION REQUIRED: `failCgroupV1` will be set to true from 1.35. 
>   This means that nodes will not start on a cgroup v1 by default. This puts cgroup v1 into a deprecated state. ([#134298](https://github.com/kubernetes/kubernetes/pull/134298), [@kannon92](https://github.com/kannon92))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Core API validation errors gain format-origin metadata

- Affects: core API object submissions (Inferred)

When a submitted core API object triggers one of the changed validation checks, its error previously lacked format-origin values; after this release, the error carries metadata such as `format=k8s-label-value`, and some existing origins use more specific `format=...` values.

Evidence: “+ allErrs = append(allErrs, field.Invalid(path.Index(valueIndex), value, msg)).WithOrigin("format=k8s-label-value")” https://github.com/kubernetes/kubernetes/pull/132825

> - Added `WithOrigin` within `apis/core/validation` with adjusted tests. ([#132825](https://github.com/kubernetes/kubernetes/pull/132825), [@PatrickLaabs](https://github.com/PatrickLaabs))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `StorageVersionMigration` API moves from `v1alpha1` to `v1beta1`

- Affects: clusters using the `StorageVersionMigration` `v1alpha1` API or containing `v1alpha1` resources

`v1alpha1` resources could be used before the upgrade; they are no longer supported afterward.

Evidence: “Action required: Any use of the v1alpha1 api is no longer supported and users must remove any v1alpha1 resources prior to upgrade.” https://github.com/kubernetes/kubernetes/pull/134784

> - Added the `StorageVersionMigration` `v1beta1` API and removed the `v1alpha1` API.
>   ACTION REQUIRED: The `v1alpha1` API is no longer supported. Users must remove any `v1alpha1` resources before upgrading. ([#134784](https://github.com/kubernetes/kubernetes/pull/134784), [@michaelasp](https://github.com/michaelasp)) [SIG API Machinery, Apps, Auth, Etcd and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `DeviceTaintRule` drops `DeviceClass` and CEL selector fields

- Affects: `DeviceTaintRule` resources with `DeviceClass` or CEL selectors

Evidence: “In addition, fields in DeviceTaintRule which cannot be supported properly without more fundamental changes (DeviceClass, CEL selectors) get removed.” https://github.com/kubernetes/kubernetes/pull/134152

> - DRA device taints: `DeviceTaintRule` status provides information about the rule, including whether Pods still need to be evicted (`EvictionInProgress` condition). The newly added `None` effect can be used to preview what a `DeviceTaintRule` would do if it used the `NoExecute` effect and to taint devices (`device health`) without immediately affecting scheduling or running Pods. ([#134152](https://github.com/kubernetes/kubernetes/pull/134152), [@pohly](https://github.com/pohly)) [SIG API Machinery, Apps, Auth, Node, Release, Scheduling and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `DynamicResourceAllocation` locked to enabled for core DRA

- Affects: setups configured to disable `DynamicResourceAllocation` (Inferred)

In v1.34, the gate was enabled by default but was not yet locked to that default.

> - DRA: The `DynamicResourceAllocation` feature gate for the core functionality (GA in `v1.34`) has now been locked to enabled-by-default and cannot be disabled anymore. ([#134452](https://github.com/kubernetes/kubernetes/pull/134452), [@pohly](https://github.com/pohly)) [SIG Auth, Node, Scheduling and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Feature gates now require enabled dependencies

- Affects: feature-gate configurations that enable features with disabled dependencies, including `AllAlpha=true`
- Action required: For configurations using `AllAlpha=true`, enable the depended-on disabled-by-default beta feature gates with `AllBeta=true` or by explicitly listing those dependencies; otherwise, the configuration is rejected.

Configurations that enable a feature while its dependency is disabled are rejected at startup, rather than failing silently.

> - Feature gate dependencies are now explicit, and validated at startup. A feature can no longer be enabled if it depends on a disabled feature. In particular, this means that `AllAlpha=true` will no longer work without enabling disabled-by-default beta features that are depended on (either with `AllBeta=true` or explicitly enumerating the disabled dependencies). ([#133697](https://github.com/kubernetes/kubernetes/pull/133697), [@tallclair](https://github.com/tallclair)) [SIG API Machinery, Architecture, Cluster Lifecycle and Node]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Default gogo protobuf integration removed from Kubernetes API Go types

- Affects: Go consumers of Kubernetes API types, especially integrations using the gogo type registry, relying on `ProtoMessage()`, or using protobuf libraries
- Removed in: v1.36

Before this release, the types used `github.com/gogo/protobuf` at runtime, were registered in its global type registry, and implemented `ProtoMessage()` by default. The marker methods can be restored for one release with the `kubernetes_protomessage_one_more_release` build tag; they are removed in `v1.36`.

> - Kubernetes API Go types removed runtime use of the `github.com/gogo/protobuf` library, and are no longer registered into the global gogo type registry. Kubernetes API Go types were not suitable for use with the `google.golang.org/protobuf` library, and no longer implement `ProtoMessage()` by default to avoid accidental incompatible use. If removal of these marker methods impacts your use, it can be re-enabled for one more release with a `kubernetes_protomessage_one_more_release` build tag, but will be removed in `v1.36`. ([#134256](https://github.com/kubernetes/kubernetes/pull/134256), [@liggitt](https://github.com/liggitt)) [SIG API Machinery, Apps, Architecture, Auth, CLI, Cluster Lifecycle, Instrumentation, Network, Node, Scheduling and Storage]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Pod observed-generation tracking promoted to GA; feature gate locked to default

- Affects: clusters using Pod observed-generation tracking or configuring the `PodObservedGenerationTracking` feature gate

The feature was beta and default-on in v1.34.

Evidence: “+ {Version: version.MustParse("1.35"), Default: true, PreRelease: featuregate.GA, LockToDefault: true}, // GA in 1.35, remove in 1.38” https://github.com/kubernetes/kubernetes/pull/134948

> - Promoted `PodObservedGenerationTracking` to GA. ([#134948](https://github.com/kubernetes/kubernetes/pull/134948), [@natasha41575](https://github.com/natasha41575)) [SIG API Machinery, Apps, Node, Scheduling and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### JobManagedBy promoted to general availability with its gate locked on

- Affects: setups configuring the `JobManagedBy` feature gate or using Jobs governed by the feature
- Removed in: not stated

The feature gate can no longer be turned off and is slated for removal in a future Kubernetes release.

> - Promoted the `JobManagedBy` feature to general availability. The `JobManagedBy` feature gate was locked to `true` and will be removed in a future Kubernetes release. ([#135080](https://github.com/kubernetes/kubernetes/pull/135080), [@dejanzele](https://github.com/dejanzele)) [SIG API Machinery, Apps and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `StrictCostEnforcementForVAP` and `StrictCostEnforcementForWebhooks` feature gates removed

- Affects: setups configured with `StrictCostEnforcementForVAP` or `StrictCostEnforcementForWebhooks` (Inferred)

Both gates had been locked on since `v1.32`. The release does not describe a change to the enforcement behavior they controlled.

> - Removed the `StrictCostEnforcementForVAP` and `StrictCostEnforcementForWebhooks` feature gates, which were locked since `v1.32`. ([#134994](https://github.com/kubernetes/kubernetes/pull/134994), [@liggitt](https://github.com/liggitt)) [SIG API Machinery, Auth, Node and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Pod Certificates API removed from `certificates` `v1alpha1`

- Affects: clients, manifests, or automation using the Pod Certificates API in `certificates` `v1alpha1` (Inferred)

The API is no longer available in `certificates` `v1alpha1`; the beta API is in `certificates` `v1beta1`.

Evidence: “This PR promotes the Pod Certificates to beta. It also removes the API from the certificates v1alpha1 API group.” https://github.com/kubernetes/kubernetes/pull/134624

> - The Pod Certificates feature moved to beta. The `PodCertificateRequest` feature gate is set disabled by default. To use the feature, users must enable the certificates API groups in `v1beta1` and enable the `PodCertificateRequest` feature gate. The `UserAnnotations` field was added to the `PodCertificateProjection` API and the corresponding `UnverifiedUserAnnotations` field was added to the `PodCertificateRequest` API. ([#134624](https://github.com/kubernetes/kubernetes/pull/134624), [@yt2985](https://github.com/yt2985)) [SIG API Machinery, Apps, Auth, Etcd, Instrumentation, Node and Testing]
> - Moved the Pod Certificates feature to beta. Added `UserAnnotations` to the `PodCertificateProjection` API and `UnverifiedUserAnnotations` to the `PodCertificateRequest` API. The `PodCertificateRequest` feature gate remains disabled by default and requires enabling the v1beta1 certificates API groups. ([#134790](https://github.com/kubernetes/kubernetes/pull/134790), [@yt2985](https://github.com/yt2985)) [SIG Auth, Instrumentation and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Service `trafficDistribution` preferences reach general availability

- Affects: Services using `PreferSameZone` or `PreferSameNode`, and clusters configuring the `PreferSameTrafficDistribution` feature gate

The values were beta in v1.34. The gate already defaulted to true and is now locked to that value.

> - The `PreferSameZone` and `PreferSameNode` values for the Service
>   `trafficDistribution` field graduated to general availability. The

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### ResourceQuota now counts device class requests in ResourceClaims

- Affects: Clusters with `DRAExtendedResource` enabled and `ResourceClaim`s that request device classes

Quota usage is based on the worst-case number of requested devices. It counts against `requests.deviceclass.resource.k8s.io/<deviceclass>` and, for classes mapped to an extended resource, `requests.<extended resource name>`.

> - Updated `ResourceQuota` to count device class requests within a `ResourceClaim` as two additional quotas when the `DRAExtendedResource` feature is enabled:
>   - `requests.deviceclass.resource.k8s.io/<deviceclass>` is charged based on the worst-case number of devices requested.
>   - Device classes mapping to an extended resource now consume `requests.<extended resource name>`. ([#134210](https://github.com/kubernetes/kubernetes/pull/134210), [@yliaog](https://github.com/yliaog)) [SIG API Machinery, Apps, Node, Scheduling and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Fine-grained supplemental groups policy reaches GA

- Affects: setups using the fine-grained supplemental groups policy or configuring its feature gate

In v1.33, the policy was beta and enabled by default. It remains enabled by default, but its feature gate is now locked to that default.

> - Graduated the fine-grained supplemental groups policy (KEP-3619) to GA. ([#135088](https://github.com/kubernetes/kubernetes/pull/135088), [@everpeace](https://github.com/everpeace)) [SIG Node and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `ControlPlaneKubeletLocalMode` locked on at GA

- Affects: kubeadm control-plane setups using `ControlPlaneKubeletLocalMode` or opting out via `/etc/kubernetes/kubelet.conf`

The gate was already enabled by default while in beta, so its default behavior is unchanged. It is now locked, and opting out requires patching the `server` field in `/etc/kubernetes/kubelet.conf`.

> - Kubeadm: Graduated the kubeadm-specific feature gate `ControlPlaneKubeletLocalMode` to GA and locked it to enabled by default. To opt out, patch the `server` field in `/etc/kubernetes/kubelet.conf`. Deprecated the subphase of `kubeadm join phase control-plane-join` called `etcd`, which is now hidden and replaced by subphase with identical functionality `etcd-join`. The `etcd` subphase will be removed in a future release. The subphase `kubelet-wait-bootstrap` of `kubeadm join` is no longer experimental and will now always run. ([#134106](https://github.com/kubernetes/kubernetes/pull/134106), [@neolit123](https://github.com/neolit123))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `kubeadm join` `etcd` subphase replaced by `etcd-join`

- Affects: scripts or operators invoking the `etcd` subphase of `kubeadm join phase control-plane-join` (Inferred)
- Removed in: not stated

Invoking `etcd` no longer runs the phase; `etcd-join` provides the same functionality.

> - Kubeadm: Graduated the kubeadm-specific feature gate `ControlPlaneKubeletLocalMode` to GA and locked it to enabled by default. To opt out, patch the `server` field in `/etc/kubernetes/kubelet.conf`. Deprecated the subphase of `kubeadm join phase control-plane-join` called `etcd`, which is now hidden and replaced by subphase with identical functionality `etcd-join`. The `etcd` subphase will be removed in a future release. The subphase `kubelet-wait-bootstrap` of `kubeadm join` is no longer experimental and will now always run. ([#134106](https://github.com/kubernetes/kubernetes/pull/134106), [@neolit123](https://github.com/neolit123))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `AllowOverwriteTerminationGracePeriodSeconds` locked to false

- Affects: Kubelet setups with a non-default `AllowOverwriteTerminationGracePeriodSeconds` setting (Inferred)

The gate was introduced in v1.32, when it could be changed from its false default, and was deprecated.

Evidence: “This feature gate was added in version v1.32, and no issues have been reported by users within three release versions. Therefore, we can lock it down and prohibit modifications.” https://github.com/kubernetes/kubernetes/pull/133792

> - Locked down the `AllowOverwriteTerminationGracePeriodSeconds` feature gate. ([#133792](https://github.com/kubernetes/kubernetes/pull/133792), [@HirazawaUi](https://github.com/HirazawaUi))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `ExecProbeTimeout` locked to true

- Affects: Kubelet configurations that set `ExecProbeTimeout` to false (Inferred)

The gate was already generally available and defaulted to true, but could previously be set to false. Configurations with that setting are no longer accepted.

Evidence: “This is a bug fix so the feature gate will be GA and on by default from the start. Timeline is to lock the feature flag to true in 1.35.” https://github.com/kubernetes/kubernetes/pull/134635

> - Locked the (generally available) feature gate `ExecProbeTimeout` to true. ([#134635](https://github.com/kubernetes/kubernetes/pull/134635), [@vivzbansal](https://github.com/vivzbansal)) [SIG Node and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `ImageGCMaximumAge` is stable, with its feature gate locked on

- Affects: Kubelet configurations using the `ImageMaximumGCAge` feature gate or `imageMaximumGCAge` setting (Inferred)

The kubelet uses the configured maximum image-GC age without requiring the feature gate. The gate was already enabled by default, so that default behavior is unchanged.

Evidence: “{Version: version.MustParse("1.30"), Default: true, PreRelease: featuregate.Beta}, + {Version: version.MustParse("1.35"), Default: true, PreRelease: featuregate.GA, LockToDefault: true}, // remove in 1.38” https://github.com/kubernetes/kubernetes/pull/134736

> - Promoted `ImageGCMaximumAge` to stable. ([#134736](https://github.com/kubernetes/kubernetes/pull/134736), [@haircommander](https://github.com/haircommander)) [SIG Node and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `InPlacePodVerticalScaling` promoted to GA with its feature gate locked on

- Affects: setups using `InPlacePodVerticalScaling`, especially those setting its feature gate to false (Inferred)

Evidence: “{Version: version.MustParse("1.33"), Default: true, PreRelease: featuregate.Beta}, + {Version: version.MustParse("1.35"), Default: true, PreRelease: featuregate.GA, LockToDefault: true}, // remove in 1.38” https://github.com/kubernetes/kubernetes/pull/134949

> - Promoted `InPlacePodVerticalScaling` to GA. ([#134949](https://github.com/kubernetes/kubernetes/pull/134949), [@natasha41575](https://github.com/natasha41575)) [SIG API Machinery, Node and Scheduling]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `kubectl` command headers no longer honor the opt-out

- Affects: `kubectl` invocations with `KUBECTL_COMMAND_HEADERS` set to false or zero (Inferred)

Previously, setting `KUBECTL_COMMAND_HEADERS` to false or zero disabled command headers.

> - Promoted `kubectl` command headers to stable. ([#134777](https://github.com/kubernetes/kubernetes/pull/134777), [@soltysh](https://github.com/soltysh)) [SIG CLI and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `EnvFiles` enabled by default as a beta feature

- Affects: Kubelet workloads using `EnvFiles` and environment files (Inferred)

In v1.34, the `EnvFiles` gate was alpha and disabled by default, and the earlier syntax was less restricted. Environment-file syntax is now limited to a subset of POSIX shell syntax, with every variable value enclosed in single quotes.

> - Promoted the `EnvFiles` feature gate to beta and is enabled by default. Additionally, the syntax specification for environment variables has been restricted to a subset of POSIX shell syntax (all variable values must be wrapped in single quotes). ([#134414](https://github.com/kubernetes/kubernetes/pull/134414), [@HirazawaUi](https://github.com/HirazawaUi)) [SIG Node and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `kubectl exec` now requires `--` before the command

- Affects: automation or commands using `kubectl exec` (Inferred)

> - Changed `kubectl exec` syntax to require `--` before the command. The form `kubectl exec [POD] [COMMAND]` is no longer supported; use `kubectl exec [POD] -- [COMMAND]` instead. ([#133841](https://github.com/kubernetes/kubernetes/pull/133841), [@yangjunmyfm192085](https://github.com/yangjunmyfm192085))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Generated `ResourceClaim` objects no longer set `BlockOwnerDeletion`

- Affects: setups using `ResourceClaimTemplate`-generated claims or scheduler-created `extendedResourceClaim` objects
- Action required: an explicit finalizer if the parent must block deletion until the child is deleted

Parent deletion is no longer blocked until these child claims are deleted; that behavior requires an explicit finalizer.

Evidence: “If the parent object must block until the child object is deleted, then use an explicit finalizer.” https://github.com/kubernetes/kubernetes/pull/134956

> - Removed `BlockOwnerDeletion` from `ResourceClaim` created from `ResourceClaimTemplate` and from `extendedResourceClaim` created by the `scheduler`. ([#134956](https://github.com/kubernetes/kubernetes/pull/134956), [@yliaog](https://github.com/yliaog)) [SIG Apps, Node and Scheduling]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `kubectl scale` missing-resource error format changed

- Affects: resource scaling with `kubectl`

When the specified resource did not exist, `kubectl scale` previously returned `error: no objects passed to scale <GroupResource> "<ResourceName>" not found`. It now returns `Error from server (NotFound): <GroupResource> "<ResourceName>" not found`, matching the format used by commands such as `kubectl get`.

Evidence: “Previously, `kubectl scale` returned the error message `error: no objects passed to scale <GroupResource> "<ResourceName>" not found` when the specified resource did not exist.” https://github.com/kubernetes/kubernetes/pull/134017

> - Updated `kubectl scale` to return a consistent error message when a specified resource is not found. Previously, it returned: `error: no objects passed to scale <GroupResource> "<ResourceName>" not found`. It now matches the format used by other commands (e.g., `kubectl get`): `Error from server (NotFound): <GroupResource> "<ResourceName>" not found`. ([#134017](https://github.com/kubernetes/kubernetes/pull/134017), [@mochizuki875](https://github.com/mochizuki875))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `kubectl describe` drops the `certificates/v1beta1` fallback

- Affects: `CertificateSigningRequest` resources served through `certificates/v1beta1` (Inferred)

Previously, `kubectl describe` tried `certificates/v1beta1` if the `certificates/v1` lookup failed. Now it returns the error from the `certificates/v1` lookup.

> - Dropped support for `certificates/v1beta1` `CertificateSigningRequest` in `kubectl`. ([#134782](https://github.com/kubernetes/kubernetes/pull/134782), [@scaliby](https://github.com/scaliby))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `kubectl describe` fallback for `EndpointSlice` removed

- Affects: `EndpointSlice` resources served through `discovery/v1beta1` (Inferred)

`kubectl describe` previously tried `discovery/v1` first and fell back to `discovery/v1beta1` if that lookup failed. Now, a failed `discovery/v1` lookup returns an error instead of falling back.

> - Dropped support for `discovery/v1beta1` `EndpointSlice` in `kubectl`. ([#134913](https://github.com/kubernetes/kubernetes/pull/134913), [@scaliby](https://github.com/scaliby))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `kubectl describe` drops `networking/v1beta1` support for `IngressClass`

- Affects: `kubectl describe` on `IngressClass` resources served through `networking/v1beta1` (Inferred)

Previously, the describer supported both `networking/v1beta1` and `networking/v1`. Now a failed `networking/v1` lookup returns an error instead of falling back to `networking/v1beta1`.

Evidence: “Dropped support for networking/v1beta1 IngressClass in kubectl” https://github.com/kubernetes/kubernetes/pull/135108

> - Dropped support for `networking/v1beta1` `Ingress` in `kubectl`. ([#135108](https://github.com/kubernetes/kubernetes/pull/135108), [@scaliby](https://github.com/scaliby))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### kubectl describe removes the Ingress `networking/v1beta1` fallback

- Affects: Ingress resources served through `networking/v1beta1` and described with `kubectl describe` (Inferred)

Previously, `kubectl describe` tried `networking/v1` first and fell back to `networking/v1beta1` if that lookup failed.

> - Dropped support for `networking/v1beta1` `Ingress` in `kubectl`. ([#135176](https://github.com/kubernetes/kubernetes/pull/135176), [@scaliby](https://github.com/scaliby))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `kubectl describe` drops `policy/v1beta1` PodDisruptionBudget fallback

- Affects: `kubectl describe` use for `PodDisruptionBudget` resources served through `policy/v1beta1` (Inferred)

Previously, a `NotFound` from the `policy/v1` lookup triggered a fallback to `policy/v1beta1`.

Evidence: “Dropped support for policy/v1beta1 PodDisruptionBudget in kubectl” https://github.com/kubernetes/kubernetes/pull/134685

> - Dropped support for `policy/v1beta1` PodDisruptionBudget in kubectl. ([#134685](https://github.com/kubernetes/kubernetes/pull/134685), [@scaliby](https://github.com/scaliby))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `HashObject` removed from the kubectl utility package

- Affects: Go code importing `k8s.io/kubectl/pkg/util` and calling `HashObject` (Inferred)

Before this release, it encoded an object and returned its MD5 hash.

Evidence: “// HashObject returns the hash of a Object hash by a Codec” https://github.com/kubernetes/kubernetes/pull/133511

> - Eliminated and prevented future use of the `md5` algorithm in favor of more appropriate hashing algorithms. ([#133511](https://github.com/kubernetes/kubernetes/pull/133511), [@BenTheElder](https://github.com/BenTheElder)) [SIG Apps, Architecture, CLI, Cluster Lifecycle, Network, Node, Security, Storage and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Kubelet removes `--pod-infra-container-image` flag

- Affects: Kubelet configurations using `--pod-infra-container-image`, including kubeadm `extraArgs`
- Action required: Remove `--pod-infra-container-image` from non-kubeadm kubelet configuration and, for kubeadm, from `extraArgs` in the kubelet configuration file.

The flag had been applied by kubeadm even though its function had moved to CRI. A retained flag can cause kubelet startup failures after the upgrade.

> - Kubeadm: stopped applying the `--pod-infra-container-image` flag for the kubelet. The flag has been deprecated and no longer served a purpose in the kubelet as the logic was migrated to CRI (Container Runtime Interface). During upgrade, kubeadm will attempt to remove the flag from the file `/var/lib/kubelet/kubeadm-flags.env`. ([#133778](https://github.com/kubernetes/kubernetes/pull/133778), [@carlory](https://github.com/carlory)) [SIG Cloud Provider and Cluster Lifecycle]
> - ACTION REQUIRED:
>   Removed the `--pod-infra-container-image` flag from `kubelet` command line. For `non-kubeadm` clusters, users must manually remove this flag from their `kubelet` configuration to prevent startup failures before upgrading `kubelet`. For `kubeadm` clusters, if users pass extra arguments to the `kubelet` like `--pod-infra-container-image`, it will be written to the `kubelet` env file during the `init` phase. `kubeadm` does not remove it during the `init` or `join` phase, so users must manually remove it from `extraArgs` in the `kubelet` configuration file. ([#133779](https://github.com/kubernetes/kubernetes/pull/133779), [@carlory](https://github.com/carlory))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Scheduler framework types moved to `k8s.io/kube-scheduler/framework`

- Affects: Go code and scheduler plugins importing the listed types from `k/k/pkg/scheduler/framework`
- Action required: Users of the moved types should update their import paths to `k8s.io/kube-scheduler/framework`.

The interfaces remain unchanged.

> - Moved Types in `k/k/pkg/scheduler/framework`:
>   `Handle`,
>   `Plugin`,
>   `PreEnqueuePlugin`, `QueueSortPlugin`, `EnqueueExtensions`, `PreFilterExtensions`, `PreFilterPlugin`, `FilterPlugin`, `PostFilterPlugin`, `PreScorePlugin`, `ScorePlugin`, `ReservePlugin`, `PreBindPlugin`, `PostBindPlugin`, `PermitPlugin`, `BindPlugin`,
>   `PodActivator`, `PodNominator`, `PluginsRunner`,
>   `LessFunc`, `ScoreExtensions`, `NodeToStatusReader`, `NodeScoreList`, `NodeScore`, `NodePluginScores`, `PluginScore`, `NominatingMode`, `NominatingInfo`, `WaitingPod`, `PreFilterResult`, `PostFilterResult`,
>   `Extender`,
>   `NodeInfoLister`, `StorageInfoLister`, `SharedLister`, `ResourceSliceLister`, `DeviceClassLister`, `ResourceClaimTracker`, `SharedDRAManager`
>   
>   to package `k8s.io/kube-scheduler/framework`. Users should update import paths. The interfaces don't change.

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Scheduler `Parallelizer` interface moved to staging package

- Affects: Go code importing the `Parallelizer` interface from `k/k/pkg/scheduler/framework/parallelism` (Inferred)
- Action required: `Parallelizer` interface imports use `k8s.io/kube-scheduler/framework`

The `Parallelizer` struct remains in its existing location in k/k.

>   Type `Parallelizer` in `k/k/pkg/scheduler/framework/parallelism` has been split into interface `Parallelizer` (in `k8s.io/kube-scheduler/framework`) and `struct Parallelizer` (location unchanged in k/k). Plugin developers should update the import path to staging repo. ([#133172](https://github.com/kubernetes/kubernetes/pull/133172), [@ania-borowiec](https://github.com/ania-borowiec)) [SIG Node, Release, Scheduling, Storage and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### DRA protocol definitions switch from gogo to standard protobuf

- Affects: Code or components using the DRA protocol definitions in `k8s.io/kubelet/pkg/apis/dra` (Inferred)

The package's old definitions used gogo; the new ones use standard `protoc`/`google.golang.org/protobuf` definitions.

Evidence: “Use standard protoc for the dra instead of gogo.” https://github.com/kubernetes/kubernetes/pull/133026

> - Removed deprecated gogo protocol definitions from `k8s.io/kubelet/pkg/apis/dra` in favor of `google.golang.org/protobuf`. ([#133026](https://github.com/kubernetes/kubernetes/pull/133026), [@saschagrunert](https://github.com/saschagrunert)) [SIG API Machinery and Node]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `SizeMemoryBackedVolumes` feature gate removed

- Affects: Kubelet configurations with the `SizeMemoryBackedVolumes` feature gate (Inferred)

The gate had been enabled by default and locked to that setting since 1.32; memory-backed `emptyDir` sizing now applies without checking it.

Evidence: “This cleans up SizeMemoryBackedVolumes feature gate after it was graduated to stable back in 1.32.” https://github.com/kubernetes/kubernetes/pull/133720

> - Removed general available feature-gate `SizeMemoryBackedVolumes`. ([#133720](https://github.com/kubernetes/kubernetes/pull/133720), [@carlory](https://github.com/carlory)) [SIG Node, Storage and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `ComponentSLIs` feature gate removed

- Affects: setups that specify the `ComponentSLIs` feature gate (Inferred)

The gate was present in the prior release; the feature had been stable since Kubernetes `v1.32`.

> - Removed the `ComponentSLIs` feature gate, as it was promoted to stable in the Kubernetes `v1.32` release. ([#133742](https://github.com/kubernetes/kubernetes/pull/133742), [@carlory](https://github.com/carlory)) [SIG Architecture and Instrumentation]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `storage.k8s.io/v1alpha1` `VolumeAttributesClass` API no longer served

- Affects: clients or manifests using `storage.k8s.io/v1alpha1` `VolumeAttributesClass` (Inferred)

Evidence: “The storage.k8s.io/v1alpha1 VolumeAttributesClass API is no longer served in 1.35” https://github.com/kubernetes/kubernetes/pull/134625

> - Removed the `VolumeAttributesClass` resource from the `storage.k8s.io/v1alpha1` API in `v1.35`. ([#134625](https://github.com/kubernetes/kubernetes/pull/134625), [@liggitt](https://github.com/liggitt)) [SIG API Machinery, Etcd, Storage and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Aggregated discovery v2beta1 type removed

- Affects: API server configurations using `AggregatedDiscoveryRemoveBetaType` or the aggregated discovery v2beta1 type (Inferred)
- Removed in: not stated

The `AggregatedDiscoveryRemoveBetaType` gate is deprecated and locked to `true`.

> - The `AggregatedDiscoveryRemoveBetaType` feature gate was deprecated and locked to `true`. ([#134230](https://github.com/kubernetes/kubernetes/pull/134230), [@Jefftree](https://github.com/Jefftree))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `SystemdWatchdog` feature gate locked to its default

- Affects: Kubelet systemd watchdog integration or `SystemdWatchdog` feature-gate configuration
- Removed in: v1.37

Kubelet systemd watchdog functionality can be enabled through systemd without configuring the feature gate.

> - The `SystemdWatchdog` feature gate has been locked to default and will be removed in future release. The systemd watchdog functionality in `kubelet` can be enabled via systemd without any feature gate configuration. See the [systemd watchdog documentation](https://kubernetes.io/docs/reference/node/systemd-watchdog/) for more information. ([#134691](https://github.com/kubernetes/kubernetes/pull/134691), [@SergeyKanzhelev](https://github.com/SergeyKanzhelev))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### kubeadm `WaitForAllControlPlaneComponents` gate removed

- Affects: kubeadm configurations that specify `WaitForAllControlPlaneComponents` (Inferred)

The gate was already locked to enabled by default in v1.34.

> - `kubeadm`: Removed the `WaitForAllControlPlaneComponents` feature gate, which graduated to GA in `v1.34` and was locked to enabled by default. ([#134781](https://github.com/kubernetes/kubernetes/pull/134781), [@neolit123](https://github.com/neolit123))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### kubeadm raises the minimum supported external etcd version

- Affects: kubeadm setups using external etcd with control plane versions `v1.32`, `v1.33`, or `v1.34`

The minimum supported external etcd version was `3.5.23-0`; it is now `3.5.24-0`.

> - `kubeadm`: Updated the supported etcd version to `v3.5.24` for control plane versions `v1.32`, `v1.33`, and `v1.34`. ([#134779](https://github.com/kubernetes/kubernetes/pull/134779), [@joshjms](https://github.com/joshjms)) [SIG API Machinery, Cloud Provider, Cluster Lifecycle, Etcd and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

## Deprecations (2)

### kube-proxy `ipvs` mode deprecated

- Affects: clusters with kube-proxy configured in `ipvs` mode (Inferred)
- Removed in: not stated

The mode remains available in this release. At startup, kube-proxy emits a deprecation warning and a Warning event with reason `IPVSDeprecation`.

> - Marked `ipvs` mode in kube-proxy as deprecated, which will be removed in a future version of Kubernetes. Users are encouraged to migrate to `nftables`. ([#134539](https://github.com/kubernetes/kubernetes/pull/134539), [@adrianmoisey](https://github.com/adrianmoisey))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `PreferClose` traffic distribution value deprecated

- Affects: Services whose `trafficDistribution` field uses `PreferClose`
- Removed in: not stated

Services using `PreferClose` receive a deprecation warning; `PreferSameZone` is the suggested replacement.

>   `PreferClose` value is now deprecated in favor of the more explicit
>   `PreferSameZone`. ([#134457](https://github.com/kubernetes/kubernetes/pull/134457), [@danwinship](https://github.com/danwinship)) [SIG API Machinery, Apps, Network and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

## Important fixes (54)

### Partitionable Devices allocation excludes devices from incomplete pools

- Affects: Partitionable Devices pools that may be incomplete

Previously, allocation could consider devices from incomplete pools when selecting devices.

> - Updated the Partitionable Devices feature to support referencing counter sets across ResourceSlices within the same resource pool. Devices from incomplete pools were no longer considered for allocation. This change introduced backwards-incompatible updates to the alpha feature, requiring any ResourceSlices using it to be removed before upgrading or downgrading between v1.34 and v1.35. ([#134189](https://github.com/kubernetes/kubernetes/pull/134189), [@mortent](https://github.com/mortent)) [SIG API Machinery, Node, Scheduling and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Scheduler scoring for DRA-backed extended resources fixed

- Affects: DRA-backed extended resources on nodes where `Allocatable` is 0 (Inferred)

Previously, requests for DRA-backed extended resources were excluded from node scoring, breaking backward compatibility for workloads migrating from device-plugin-backed resources. The scheduler now uses available and requested device quantities to score them.

> - Implemented scoring for DRA-backed extended resources. ([#134058](https://github.com/kubernetes/kubernetes/pull/134058), [@bart0sh](https://github.com/bart0sh)) [SIG Node, Scheduling and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Kubelet sandbox ordering fixed for clock rollback

- Affects: Kubelet Pod sandbox ordering (Inferred)

System-clock rollback, especially during startup when NTP starts around static Pods, caused the ordering issue. Ordering now uses the sandbox attempt count when metadata is available, and creation time if either sandbox's metadata is nil.

> - Updated sandbox ordering to use by attempt count or creation time. ([#130551](https://github.com/kubernetes/kubernetes/pull/130551), [@yylt](https://github.com/yylt))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### SELinux warning controller emits previously missed conflict events

- Affects: clusters using the SELinux warning controller (Inferred)

Previously, some SELinux label conflicts produced no warning events. The comparison now includes the final label component (level and categories).

> - Fixed SELinux warning controller not emitting events on some SELinux label conflicts. ([#133425](https://github.com/kubernetes/kubernetes/pull/133425), [@jsafrane](https://github.com/jsafrane)) [SIG Apps, Storage and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### [Silent failure] ReplicaCalculator replica-count overflow fixed

- Affects: setups using `ReplicaCalculator`

When a computed replica count exceeded `int32.MaxValue`, integer overflow could wrap it to a small value and cause incorrect scaling decisions. The count is now capped at `int32.MaxValue` instead of wrapping.

> - Fixed `replicaCount` calculation exceeding max `int32`. ([#126979](https://github.com/kubernetes/kubernetes/pull/126979), [@omerap12](https://github.com/omerap12)) [SIG Apps and Autoscaling]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Windows kube-proxy stale `RemoteEndpoints` cleanup fixed

- Affects: Windows clusters using kube-proxy (winkernel) with Deployments referenced by multiple Services

When multiple Services were updated during one `syncProxyRules()` call, premature clearing of the `terminatedEndpoints` map could leave stale `RemoteEndpoints`. The map now remains available throughout the batch of updates and is cleared once at the start of the sync.

> - Fixed a Windows kube-proxy (winkernel) issue where stale `RemoteEndpoints`
>   remained when a Deployment was referenced by multiple Services due to premature
>   clearing of the `terminatedEndpoints` map. ([#135146](https://github.com/kubernetes/kubernetes/pull/135146), [@princepereira](https://github.com/princepereira)) [SIG Network and Windows]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Out-of-range IPv6 `ServiceCIDR` allocations fixed

- Affects: clusters allocating addresses from 64-bit IPv6 `ServiceCIDRs`

When an `ipAllocator` uses a 64-bit IPv6 CIDR, an improper cast in `addOffsetAddress` could cause it to allocate addresses outside the CIDR range.

> - Fixed a bug where 64-bit IPv6 `ServiceCIDRs` allocated addresses outside the subnet range. ([#134193](https://github.com/kubernetes/kubernetes/pull/134193), [@hoskeri](https://github.com/hoskeri))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Deleted Pod capacity accounting fixed in `kube-scheduler`

- Affects: clusters using `kube-scheduler`

When a Pod was deleted during the binding phase, it remained in the cache and counted against node capacity, so higher-priority Pods could repeatedly try to preempt it, wasting scheduling cycles and API calls. Once the scheduler registers the deletion event, the Pod no longer counts against node capacity, though its binding cycle can still fail as before.

> - Fixed a bug where a deleted Pod in the binding phase continued to occupy space on the node in `kube-scheduler`. ([#134157](https://github.com/kubernetes/kubernetes/pull/134157), [@macsko](https://github.com/macsko)) [SIG Scheduling and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### DRA resource health reporting in Pod status fixed

- Affects: DRA resource claims in Pods

When a claim was generated from a `ResourceClaimTemplate` or its local name in the Pod spec differed from the actual `ResourceClaim` object's name, the kubelet could omit device health status from Pod status.

> - Fixed a bug where the health of a DRA resource was not reported in the Pod status if the resource claim was generated from a template or used a different local name in the Pod spec. ([#134875](https://github.com/kubernetes/kubernetes/pull/134875), [@Jpsassine](https://github.com/Jpsassine)) [SIG Node and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### ResourceClaim config request references fixed

- Affects: ResourceClaims with config `FromClass` entries

Previously, config entries could omit which request they belonged to, making it difficult for a DRA driver to match configs to requests; two requests using the same `DeviceClass` could also duplicate a config `FromClass`. The entries now reference their corresponding requests, including requests that share a `DeviceClass`.

> - Fixed an issue where requests for a config `FromClass` in the `ResourceClaim` status were not referenced. ([#134793](https://github.com/kubernetes/kubernetes/pull/134793), [@LionelJouin](https://github.com/LionelJouin))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### [Silent failure] `kubelet` `/configz` cgroup driver reporting fixed

- Affects: kubelets using the container runtime for the cgroup driver setting

In this situation, `/configz` reported an incorrect `kubeletconfig.cgroupDriver` value. It now reports the setting received from the container runtime.

> - Fixed an issue where the `kubelet` `/configz` endpoint reported an incorrect value for `kubeletconfig.cgroupDriver` when the cgroup driver setting was received from the container runtime. ([#134743](https://github.com/kubernetes/kubernetes/pull/134743), [@marquiz](https://github.com/marquiz))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Default `serviceCIDR` controller event logging restored

- Affects: clusters running the default `serviceCIDR` controller

During initialization, the event broadcaster was shut down, preventing the controller from logging events. It now remains active until the controller context is done, allowing events to be logged.

> - Fixed an issue where the default `serviceCIDR` controller did not log events because the event broadcaster was shutdown during initialization. ([#133338](https://github.com/kubernetes/kubernetes/pull/133338), [@aojea](https://github.com/aojea))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Kubernetes API resource shell completion fixed

- Affects: shell completion for Kubernetes API resource names in `kubectl`

In v1.34, tab completion could insert a tabular row instead of the resource name. It now returns the resource name, such as `addons.k3s.cattle.io`.

> - Fixed broken shell completion for API resources. ([#133771](https://github.com/kubernetes/kubernetes/pull/133771), [@marckhouzam](https://github.com/marckhouzam))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Preemptor scheduling loop during victim deletion fixed

- Affects: clusters using pod preemption

If the preemptor was moved to the active queue while its victim was still being deleted, such as while the victim was binding, it could retry scheduling in an endless loop. It now stays unschedulable until the victim has finished being deleted.

> - Fixed incorrect behavior of preemptor pod when preemption of the victim takes long to complete. The preemptor pod should not be circling in scheduling cycles until preemption is finished. ([#134294](https://github.com/kubernetes/kubernetes/pull/134294), [@ania-borowiec](https://github.com/ania-borowiec)) [SIG Scheduling and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Kubelet `kubelet_volume_stats_*` metrics restored

- Affects: setups scraping kubelet volume statistics metrics (Inferred)

In v1.34, when DRA also called `Register` in `NewContainerManager()`, the volume stats collector was ignored and the metrics went missing.

> - Fixed missing `kubelet_volume_stats_*` metrics. ([#133890](https://github.com/kubernetes/kubernetes/pull/133890), [@huww98](https://github.com/huww98)) [SIG Instrumentation and Node]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Kubelet now honors `userNamespaces.idsPerPod`

- Affects: Kubelets configured with `userNamespaces.idsPerPod`

When `userNamespaces.idsPerPod` was configured, the kubelet ignored it and used the default value instead.

> - Fixed the kubelet to honor the `userNamespaces.idsPerPod` configuration, which was previously ignored. ([#133373](https://github.com/kubernetes/kubernetes/pull/133373), [@AkihiroSuda](https://github.com/AkihiroSuda)) [SIG Node and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Kube-controller-manager controller matching fix

- Affects: pods managed by kube-controller-manager controllers

When a pod's `ownerReference` has an incorrect UID, UID-only indexing could cause controller confusion. Matching now also accounts for the reference's namespace, kind, and name.

> - Kube-controller-manager: Resolved potential issues handling pods with incorrect uids in their `ownerReference`. ([#134654](https://github.com/kubernetes/kubernetes/pull/134654), [@liggitt](https://github.com/liggitt))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Kubeadm upgrade dry-run honors configured kubelet output directory

- Affects: Kubeadm upgrade dry-runs with `KUBEADM_UPGRADE_DRYRUN_DIR` set

Previously, kubelet configuration files could be written to the normal kubelet directory instead of the configured dry-run directory. They are now written to the configured dry-run directory.

> - Kubeadm: Fixed `KUBEADM_UPGRADE_DRYRUN_DIR` not honored in upgrade phase when writing kubelet config files. ([#134007](https://github.com/kubernetes/kubernetes/pull/134007), [@carlory](https://github.com/carlory))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### kubeadm respects the `v1beta3` control-plane timeout

- Affects: kubeadm `init` configurations using `v1beta3` and setting `APIServer.TimeoutForControlPlane` when `v1beta4` is the default

Previously, kubeadm used the default `4m0s` instead of the supplied timeout for the control-plane health check. It now uses the supplied value.

> - Kubeadm: Fixed a bug where `ClusterConfiguration.APIServer.TimeoutForControlPlane` from `v1beta3` was not respected in newer kubeadm versions where `v1beta4` is the default. ([#133513](https://github.com/kubernetes/kubernetes/pull/133513), [@tom1299](https://github.com/tom1299))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Namespace included in `kubectl autoscale` client dry-run output

- Affects: use of `kubectl autoscale` with `--dry-run=client` (Inferred)

Previously, the printed HPA object lacked the namespace and was invalid.

> - Namespace is now included in the `--dry-run=client` output for `HorizontalPodAutoscaler (HPA)` objects. ([#134263](https://github.com/kubernetes/kubernetes/pull/134263), [@ardaguclu](https://github.com/ardaguclu)) [SIG CLI and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Non-positive `log-flush-frequency` values rejected

- Affects: components configured with `log-flush-frequency` set to zero or less

Previously, these values passed validation and could cause a panic. They now return a validation error instead.

> - Added validation to ensure `log-flush-frequency` is a positive value, returning an error instead of causing a panic. ([#133540](https://github.com/kubernetes/kubernetes/pull/133540), [@BenTheElder](https://github.com/BenTheElder)) [SIG Architecture, Instrumentation, Network and Node]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Kube-apiserver validation panic fixed

- Affects: custom resources whose CRDs declare a status subresource but omit `status` from `openAPIV3Schema`

Previously, validating such a custom resource could panic; validation no longer causes that possible panic.

> - Kube-apiserver: Fixed a possible panic validating a custom resource whose `CustomResourceDefinition` indicates a status subresource exists, but which does not define a `status` property in the `openAPIV3Schema`. ([#133721](https://github.com/kubernetes/kubernetes/pull/133721), [@fusida](https://github.com/fusida)) [SIG API Machinery, Apps, Architecture, Auth, Autoscaling, CLI, Cloud Provider, Cluster Lifecycle, Etcd, Instrumentation, Network, Node, Release, Scheduling, Storage and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### CSI driver requirements enforced during Pod scheduling

- Affects: Pods that require a CSI driver

These Pods are not scheduled onto nodes that lack the required driver.

> - Prevented Pods from being scheduled onto nodes that lack the required CSI driver. ([#135012](https://github.com/kubernetes/kubernetes/pull/135012), [@gnufied](https://github.com/gnufied)) [SIG API Machinery, Scheduling, Storage and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Remote runtime and image APIs gain `Close()` methods

- Affects: remote runtime and remote image APIs (Inferred)

Callers can invoke `Close()` to shut down the internal gRPC connection and avoid leaking resources.

> - Added remote runtime and image `Close()` method to be able to close the connection. ([#133211](https://github.com/kubernetes/kubernetes/pull/133211), [@saschagrunert](https://github.com/saschagrunert)) [SIG Node]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Support for using one `PersistentVolumeClaim` across Pod volumes

- Affects: Pods that reference the same `PersistentVolumeClaim` from multiple volumes

Previously, a Pod in this configuration could remain stuck at `ContainerCreating` without an error message.

> - Added support for Pods to reference the same `PersistentVolumeClaim` across multiple volumes. ([#122140](https://github.com/kubernetes/kubernetes/pull/122140), [@huww98](https://github.com/huww98)) [SIG Node, Storage and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### DRA Device Taint `NoExecute` eviction fix

- Affects: Pods using DRA Device Taints with a `NoExecute` toleration

The scheduler did not inform the eviction controller about the toleration, so a scheduled Pod with a `NoExecute` toleration for a DRA Device Taint was evicted almost immediately.

> - DRA Device Taints: Fixed toleration of `NoExecute`. Prior to this enhancement, tolerating a `NoExecute` did not work because the scheduler did not inform the eviction controller about the toleration, so the scheduled pod got evicted almost immediately. ([#134479](https://github.com/kubernetes/kubernetes/pull/134479), [@pohly](https://github.com/pohly)) [SIG Apps, Node, Scheduling and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Automatic `ResourceClaim` reuse of extended resources

- Affects: Pods with extended resources requested by `initContainers` and allocated using an automatic `ResourceClaim`

When a later sidecar `initContainer` or regular container requests the same resource, the scheduler can reuse it when possible, reducing the Pod's total requested devices.

> - Extended resources requested by initContainers which are allocated using an automatic ResourceClaim now match the behavior of legacy device plugins, reusing the same resources requested by later sidecar initContainers or regular containers when possible, to minimize the total number of devices requested by the pod. ([#134882](https://github.com/kubernetes/kubernetes/pull/134882), [@yliaog](https://github.com/yliaog)) [SIG Apps, CLI, Node, Scheduling and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `additionalProperties: true` schema crash fixed in kube-controller-manager

- Affects: `ValidatingAdmissionPolicy` schemas with `additionalProperties: true`

Previously, such a schema could trigger a nil pointer exception in kube-controller-manager; it is now processed without that crash.

> - Fixed a bug in `ValidatingAdmissionPolicy` where schemas with `additionalProperties: true` could cause the kube-controller-manager to crash with a nil pointer exception. ([#135155](https://github.com/kubernetes/kubernetes/pull/135155), [@jpbetz](https://github.com/jpbetz))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### kube-proxy `nftables` name-based interface match fix

- Affects: clusters using kube-proxy `nftables` mode

Name-based interface matches used `iif` instead of `iifname`, so kube-proxy could not determine whether traffic came from a local source on the node, and the incorrect match could cause nftables sync failures. They now use `iifname`, fixing both issues.

> - Fixed a bug in `kube-proxy` `nftables` mode (GA as of `v1.33`) which fails to determine if traffic originates from a local source on the node. The issue was caused by using the wrong meta `iif` instead of `iifname` for name based matches. ([#134024](https://github.com/kubernetes/kubernetes/pull/134024), [@jack4it](https://github.com/jack4it))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### API server webhook Content-Type mismatch fixed

- Affects: setups with the alpha `ClientsPreferCBOR` feature gate enabled and authorization, token authentication, imagepolicy admission, or audit webhooks

When an affected webhook negotiated based on `Content-Type`, its request body was encoded as JSON while the header could name a different media type, preventing the webhook from decoding it.

> - Fixed a bug that caused apiservers to send an inappropriate Content-Type request header to authorization, token authentication, imagepolicy admission, and audit webhooks when the alpha client-go feature gate "ClientsPreferCBOR" is enabled. ([#132960](https://github.com/kubernetes/kubernetes/pull/132960), [@benluddy](https://github.com/benluddy)) [SIG API Machinery and Node]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### DRA allocation of previously used devices fixed

- Affects: setups with both `DRAConsumableCapacity` and `DRAPartitionableDevices` enabled

Allocating a device that had previously consumed the `CounterSet` could be prevented. The allocator can now use that device when its remaining capacity meets the request, including when a `CounterSet` limit applies.

> - Fixed a bug that prevented allocating the same device that was previously consuming the `CounterSet` when both `DRAConsumableCapacity` and `DRAPartitionableDevices` were enabled. ([#134103](https://github.com/kubernetes/kubernetes/pull/134103), [@sunya-ch](https://github.com/sunya-ch))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### DRAConsumableCapacity scheduling blockage fixed

- Affects: setups using the `DRAConsumableCapacity` feature

When this feature was in use, a bug could prevent the scheduler from scheduling the next pod. The scheduler no longer encounters that blocker.

> - Fixed a bug that prevents scheduling the next pod when using the `DRAConsumableCapacity` feature. ([#133706](https://github.com/kubernetes/kubernetes/pull/133706), [@sunya-ch](https://github.com/sunya-ch))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Nested-map client-side patch segmentation fault fixed

- Affects: client-side patches involving nested maps (Inferred)

When a client-side patch involving nested maps updated deeply nested JSON fields, it could panic with a segmentation fault.

> - Fixed a bug to prevent segmentation fault from occurring when updating deeply nested JSON fields. ([#134381](https://github.com/kubernetes/kubernetes/pull/134381), [@kon-angelo](https://github.com/kon-angelo)) [SIG API Machinery and CLI]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Job status update failures after resuming suspended Jobs fixed

- Affects: Jobs resumed after being started and suspended

After such a Job was resumed, its status update could fail with `status.startTime: Required value: startTime cannot be removed for unsuspended job`.

> - Fixed a bug where Job status updates fail after resuming a Job that was previously started and suspended.
>   The error message was: `status.startTime: Required value: startTime cannot be removed for unsuspended job`. ([#134769](https://github.com/kubernetes/kubernetes/pull/134769), [@dejanzele](https://github.com/dejanzele)) [SIG Apps and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### DRA `AllocationMode: All` allocation fixed

- Affects: DRA resource pools using `AllocationMode: All`

When a pool contained `ResourceSlices` that did not target the current node, it could be considered incomplete and the allocation request would not succeed. Such requests can now succeed despite those slices.

> - Fixed a bug where `AllocationMode: All` would not succeed if a resource pool contained `ResourceSlices` that were not targeting the current node. ([#134466](https://github.com/kubernetes/kubernetes/pull/134466), [@mortent](https://github.com/mortent))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Kubelet stale-cache node-affinity rejections fixed

- Affects: kubelet admission of scheduled Pods with node-affinity constraints

When the informer cache was stale after a Pod had been scheduled to a node, kubelet could reject it with `NodeAffinityFailed`. Kubelet now synchronously fetches the Node from the API server, updates its cache, and retries instead of immediately rejecting the Pod.

> - Fixed a long-standing issue where `kubelet` rejected Pods with `NodeAffinityFailed` due to a stale informer cache. ([#134445](https://github.com/kubernetes/kubernetes/pull/134445), [@natasha41575](https://github.com/natasha41575))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `kubectl api-resources` panic fixed

- Affects: use of `kubectl api-resources`

When the Discovery Client failed—including when no Kubernetes server could be reached—the command could panic. It now returns an error instead.

> - Fixed a panic in `kubectl api-resources` that occurred when the Discovery Client failed. ([#134833](https://github.com/kubernetes/kubernetes/pull/134833), [@rikatz](https://github.com/rikatz))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Admission policy namespace lookup error fixed

- Affects: Default `v1.30+` configurations using `ValidatingAdmissionPolicy` or `MutatingAdmissionPolicy` for namespaced objects in newly created namespaces

A stale namespace lister could cause a spurious `namespace not found` error when an admission policy intercepted a namespaced object in a newly created namespace. If the lister lacks that namespace, admission now looks up the live namespace object instead.

> - Fixed a spurious `namespace not found` error in default `v1.30+` configurations when using `ValidatingAdmissionPolicy` or `MutatingAdmissionPolicy` to intercept namespaced objects in newly-created namespaces. ([#135359](https://github.com/kubernetes/kubernetes/pull/135359), [@liggitt](https://github.com/liggitt))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Main-container startup stall from sidecar probe race fixed

- Affects: Pods with `restartPolicy=Never` and sidecars with startup probes

When a sidecar's startup probe initially failed and then succeeded after restart, the main container could remain stuck in `Initializing`.

> - Fixed a startup probe race condition that caused main containers to remain stuck in "Initializing" state when sidecar containers with startup probes had failed initially but succeeded on restart in pods with `restartPolicy=Never`. ([#133072](https://github.com/kubernetes/kubernetes/pull/133072), [@AadiDev005](https://github.com/AadiDev005)) [SIG Node and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Scheduler prevents duplicate asynchronous preemption requests

- Affects: workloads using scheduler asynchronous preemption

When a Pod already has a victim preemption in progress, repeated requests could leave it stuck in the unschedulable queue. The scheduler now waits for the ongoing preemption to complete before returning the Pod to the active queue to try another node.

> - Fixed an issue in asynchronous preemption: Scheduler now checks if preemption is ongoing for a Pod before initiating new preemption calls. ([#134730](https://github.com/kubernetes/kubernetes/pull/134730), [@ania-borowiec](https://github.com/ania-borowiec)) [SIG Scheduling and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Static Pod restart policies and rules now apply

- Affects: Static Pods configured with restart policies or restart rules

Previously, these settings were not applied to Static Pods because decoding used empty validation options, which also disallowed restart rules.

> - Fixed an issue that prevented restart policies and restart rules from being applied to static pods. ([#135031](https://github.com/kubernetes/kubernetes/pull/135031), [@yuanwang04](https://github.com/yuanwang04))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Extended resource claims deallocated when pods complete

- Affects: pods with extended resource claims

Previously, extended resource claims were not cleaned up when a pod completed.

> - Fixed resource claims deallocation for extended resource when Pod completes. ([#134312](https://github.com/kubernetes/kubernetes/pull/134312), [@alaypatel07](https://github.com/alaypatel07)) [SIG Apps, Node and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Kubeadm `cluster-info` context validation prevents panics

- Affects: `cluster-info` ConfigMaps with malformed kubeconfigs lacking a valid current context

When Kubeadm retrieved cluster information from such a ConfigMap, it could panic. It now reports validation errors instead.

> - Kubeadm: Added missing cluster-info context validation to prevent panics when the user has a malformed kubeconfig in the cluster-info ConfigMap that excludes a valid current context. ([#134715](https://github.com/kubernetes/kubernetes/pull/134715), [@neolit123](https://github.com/neolit123))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Kubeadm wait-control-plane waiter now connects directly to the local API server

- Affects: kubeadm `init` and `join` operations

During the `wait-control-plane` phase, the waiter previously used a client from `admin.conf` that pointed to the control-plane endpoint instead of the local API server.

> - Kubeadm: Ensured waiting for `apiserver` uses a local client that doesn't reach to the control plane endpoint and instead reaches directly to the local API server endpoint. ([#134265](https://github.com/kubernetes/kubernetes/pull/134265), [@neolit123](https://github.com/neolit123))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Kubeadm node registration lookup fixed during upgrades

- Affects: `kubeadm upgrade node` on nodes whose node name differs from the host name

Node registration information could be fetched incorrectly, resulting in an incorrect node name. It is now fetched correctly, including when the names differ.

> - Kubeadm: Fixed a bug where the node registration information for a given node was not fetched correctly during `kubeadm upgrade node` and the node name can end up being incorrect in cases where the node name is not the same as the host name. ([#134319](https://github.com/kubernetes/kubernetes/pull/134319), [@neolit123](https://github.com/neolit123))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Kubelet concurrent map write error fixed

- Affects: Kubelet setups with `LocalStorageCapacityIsolationFSQuotaMonitoring` enabled

Creating a pod with an empty volume could trigger a concurrent map write error in Kubelet.

> - Kubelet: Fixed a concurrent map write error when creating a pod with an empty volume while the `LocalStorageCapacityIsolationFSQuotaMonitoring` feature gate is enabled. ([#135174](https://github.com/kubernetes/kubernetes/pull/135174), [@carlory](https://github.com/carlory))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Kubelet DRA driver reconnect deadlock fixed

- Affects: Kubelet setups using a DRA driver with an idle gRPC connection

When the gRPC connection to a DRA driver went idle, by default after 30 minutes, kubelet's reconnect path could deadlock and leave the connection unusable.

> - Kubelet: Fixed an internal deadlock that caused the connection to a DRA driver to become unusable after being idle for 30 minutes. ([#133926](https://github.com/kubernetes/kubernetes/pull/133926), [@pohly](https://github.com/pohly))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Node and Pod Events now populate `involvedObject.apiVersion`

- Affects: Events for Nodes or Pods

> - Populated `involvedObject.apiVersion` on Events created for Nodes and Pods. ([#134545](https://github.com/kubernetes/kubernetes/pull/134545), [@novahe](https://github.com/novahe)) [SIG Cloud Provider, Network, Node, Scalability and Testing]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### Device Manager health check no longer causes watchdog kills

- Affects: kubelets using the Device Manager

Until it began listening on its port, the Device Manager was treated as unhealthy, so slow container runtime initialization could cause the watchdog to kill kubelet.

> - Slow container runtime initialization no longer causes the System WatchDog to kill the kubelet. The Device Manager was treated as unhealthy until it began listening on its port. ([#135153](https://github.com/kubernetes/kubernetes/pull/135153), [@SergeyKanzhelev](https://github.com/SergeyKanzhelev))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `kube-controller-manager` fix for spurious StatefulSet rollouts

- Affects: clusters running `kube-controller-manager` with existing StatefulSets

During a control-plane upgrade from `v1.33` to `v1.34`, a regression could trigger spurious rollouts of existing StatefulSets. With `StatefulSetSemanticRevisionComparison` enabled, the regression no longer triggers those rollouts.

> - `kube-controller-manager`: Fixed a `v1.34` regression that triggered a spurious rollout of existing StatefulSets when upgrading the control plane from `v1.33` to `v1.34`. This fix is guarded by the `StatefulSetSemanticRevisionComparison` feature gate, which is enabled by default. ([#135017](https://github.com/kubernetes/kubernetes/pull/135017), [@liggitt](https://github.com/liggitt))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `MutatingAdmissionPolicy` duplicate-list handling fixed

- Affects: `MutatingAdmissionPolicy` with `ApplyConfiguration`

Applying the policy to an existing object with duplicate list items, such as duplicate environment variables, previously failed with a Structured Merge Diff conversion error. It can now process that object, while newly generated patches remain subject to strict Structured Merge Diff enforcement.

> - Fixes a bug where `MutatingAdmissionPolicy` would fail to apply to objects with duplicate list items (like env vars). ([#135560](https://github.com/kubernetes/kubernetes/pull/135560), [@lalitc375](https://github.com/lalitc375) [SIG API Machinery]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### `k8s.io/client-go` informer `Transformer` regression fixed

- Affects: `k8s.io/client-go` setups using informers with configured `Transformer` functions

In 1.34+, informers did not use their configured `Transformer` functions because of a regression.

> - K8s.io/client-go: Fixes a regression in 1.34+ which prevented informers from using configured Transformer functions. ([#135580](https://github.com/kubernetes/kubernetes/pull/135580), [@serathius](https://github.com/serathius) [SIG API Machinery]

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### kube-apiserver returns HTTP 400 for invalid `DeleteOptions`

- Affects: clients sending invalid `DeleteOptions` in requests handled by kube-apiserver (Inferred)

When kube-apiserver received invalid `DeleteOptions`, it returned HTTP 500 instead of HTTP 400.

> - Kube-apiserver: Fixed an issue where passing invalid `DeleteOptions` incorrectly returned a 500 status instead of 400. ([#133358](https://github.com/kubernetes/kubernetes/pull/133358), [@ostrain](https://github.com/ostrain))

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

### kube-apiserver high CPU bug fixed for malformed Services

- Affects: setups using the new Cluster IP allocators with `MultiCIDRServiceAllocator` enabled

A malformed Service without a name could drive kube-apiserver CPU usage high. This malformed-Service case no longer causes the high-CPU bug.

> - Fix a bug in the kube-apiserver where a malformed Service without name can cause high CPU usage. The bug is present on the new Cluster IP allocators enabled with the feature MultiCIDRServiceAllocator (enabled by default since 1.33)

https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.35.md

## Enhancements (51)

- Observed generation field for `CustomResourceDefinition` conditions (To use: Enable the `CRDObservedGenerationTracking` feature gate; it is Beta and disabled by default in v1.35.)
- Minimum control-plane compatibility version flag
- Restart all containers on a matching container exit
- New optional kuberc fields for credential plugin allowlisting
- DeviceTaintRule status reports rule information
- Preview-only `None` effect for `DeviceTaintRule`
- `kubectl get -o kyaml` enabled by default
- Alpha in-place resizing for pod-level resources (To use: The `InPlacePodLevelResourcesVerticalScaling` feature gate, which is alpha and disabled by default in v1.35, plus `InPlacePodVerticalScaling`, `PodLevelResources`, and `NodeDeclaredFeatures`.)
- Aggregated discovery includes peer API resources (To use: `UnknownVersionInteroperabilityProxy` feature gate enabled)
- Numeric `Gt` and `Lt` toleration operators (To use: Enable the alpha `TaintTolerationComparisonOperators` feature gate, which defaults to false in v1.35.)
- Configurable OpenAPI model names for generated API types (To use: `+k8s:openapi-model-package` declaration to specify model package names)
- Constrained impersonation support (To use: Enable the `ConstrainedImpersonation` alpha feature gate; it is disabled by default in v1.35.)
- Structured JSON responses from `/statusz` (To use: `Accept: application/json;v=v1alpha1;g=config.k8s.io;as=Statusz`)
- Structured `v1alpha1` responses for `/flagz` (To use: `Accept: application/json;v=v1alpha1;g=config.k8s.io;as=Flagz` header)
- All-or-nothing scheduling with the GangScheduling kube-scheduler plugin (To use: The `GangScheduling` feature gate enabled; policy expressed through the `scheduling.k8s.io/v1alpha1` Workload API)
- Node Declared Features capability introduced (To use: `NodeDeclaredFeatures` feature gate (alpha, disabled by default in v1.35))
- Workload API and Pod reference enable workload-level scheduling
- Suspended Jobs can change scheduling directives (To use: `MutableSchedulingDirectivesForSuspendedJobs` feature gate; disabled by default)
- Mutable Persistent Volume node affinity (To use: `MutablePVNodeAffinity` feature gate, disabled by default in v1.35.0)
- Configurable DynamicResources `PreBind` wait timeout (To use: Both `DRADeviceBindingConditions` and `DRAResourceClaimDeviceStatus` feature gates enabled)
- Pod Certificates feature moves to beta (To use: Enable the certificates API groups in `v1beta1` and the `PodCertificateRequest` feature gate, which is disabled by default.)
- `UserAnnotations` and `UnverifiedUserAnnotations` added to pod certificate APIs (To use: Enable the certificates API groups in `v1beta1` and the `PodCertificateRequest` feature gate)
- Partitionable Devices support cross-`ResourceSlice` counter-set references
- DNS name format tags added for `validation-gen`
- New `kubectl` commands for viewing and updating the `kuberc` file (To use: `kubectl alpha kuberc view` and `kubectl alpha kuberc set`)
- Kubelet `statusz` endpoint lists registered paths (To use: `ComponentStatusz` feature gate)
- Resource claim metric gains a `source` label
- New `scheduler_resourceclaim_creates_total` metric for `DRAExtendedResource` scheduling (To use: The `DRAExtendedResource` feature gate must be enabled.)
- Kubelet image-manager request counter metric added
- Configurable per-device health check timeouts for DRA (To use: A DRA driver specifies a custom timeout for an individual device through the gRPC health API.)
- StatefulSet `MaxUnavailable` metrics added
- Trace profiling in `kubectl` (To use: `kubectl --profile=trace`)
- UUID format validation in `validation-gen` (To use: A string field tagged with `// +k8s:format=uuid`)
- `kubectl config set-context` gains `-n` namespace shorthand
- Watch-based route reconciliation in cloud-controller-manager (To use: Enable the `CloudControllerManagerWatchBasedRoutesReconciliation` feature gate; it is alpha and disabled by default in v1.35.)
- User namespaces for `hostNetwork` pods (To use: Enable the `UserNamespacesHostNetworkSupport` feature gate; it is disabled by default.)
- `ContainerRestartRules` promoted to beta and enabled by default (To use: `ContainerRestartRules` feature gate, enabled by default in v1.35.0)
- Kubelet image pull metrics added
- `kubectl` adds persistent `--as-user-extra` impersonation flag (To use: `--as-user-extra`)
- Resource version comparison helper in `k8s.io/apimachinery`
- Resource updates for suspended Jobs (To use: `MutablePodResourcesForSuspendedJobs` feature gate (alpha in Kubernetes v1.35; disabled by default))
- kubeadm retry errors logged during control-plane waits (To use: klog verbosity level 5)
- Separate HTTP endpoints for external etcd in kubeadm (To use: `v1beta4` `ClusterConfiguration.Etcd.ExternalEtcd`)
- `resource.k8s.io` validation migrated to declarative validation (To use: `DeclarativeValidation` for mismatch reporting through metrics; `DeclarativeValidationTakeover` for declarative validation to become the primary source of errors for migrated fields.)
- Pod Certificates beta adds user annotation fields (To use: Enable the v1beta1 certificates API groups and the `PodCertificateRequest` feature gate to use `PodCertificateRequest`.)
- JWKS fetch metrics added to the JWT authenticator (To use: `StructuredAuthenticationConfiguration` feature gate)
- `applyconfiguration-gen` generates extract functions for all subresources
- `ShareID` support in the Kubelet Plugin API (To use: Enable the `DRAConsumableCapacity` feature gate (`DRAConsumableCapacity=true`).)
- DRA device requests support implicit extended resource names
- `MutatingAdmissionPolicy` consolidates duplicate map-list entries
- Exposed endpoints listed on kube-proxy `/statusz` (To use: `ComponentStatusz` must be enabled.)

## Dependency updates (99)

99 more, none mentioning security

- `` - Updated the Kubernetes build to use Go `1.25.4`. ([#135187](https://github.com/kubernetes/kubernetes/pull/135187), [@BenTheElder](https://github.com/BenTheElder)) ``
- `` - Updated underlying images and dependencies to be compatible with Go version`1.25.3`. ([#134611](https://github.com/kubernetes/kubernetes/pull/134611), [@cpanato](https://github.com/cpanato)) [SIG Architecture, Cloud Provider, Etcd, Release, Storage and Testing] ``
- `` - Kubernetes is now built using Go `1.25.5`. ([#135609](https://github.com/kubernetes/kubernetes/pull/135609), [@cpanato](https://github.com/cpanato)) [SIG Release and Testing] ``
- `` - Bumped addon manager to use `kubectl` version `v1.32.2`. ([#130548](https://github.com/kubernetes/kubernetes/pull/130548), [@Jefftree](https://github.com/Jefftree)) [SIG Cloud Provider, Scalability and Testing] ``
- `- Updated CNI plugins to v1.8.0. ([#133837](https://github.com/kubernetes/kubernetes/pull/133837), [@saschagrunert](https://github.com/saschagrunert)) [SIG Cloud Provider, Node and Testing]`
- `` - Updated `etcd` to `v3.6.5`. ([#134251](https://github.com/kubernetes/kubernetes/pull/134251), [@joshjms](https://github.com/joshjms)) [SIG API Machinery, Cloud Provider, Cluster Lifecycle, Etcd and Testing] ``
- `- Updated cri-tools to v1.34.0. ([#133636](https://github.com/kubernetes/kubernetes/pull/133636), [@saschagrunert](https://github.com/saschagrunert)) [SIG Cloud Provider]`
- `` - Updated the Go version of Kubernetes to `1.25.3`. ([#134598](https://github.com/kubernetes/kubernetes/pull/134598), [@BenTheElder](https://github.com/BenTheElder)) ``
- `` - Updated the etcd client library to `v3.6.5`. ([#134780](https://github.com/kubernetes/kubernetes/pull/134780), [@joshjms](https://github.com/joshjms)) [SIG API Machinery, Architecture, Auth, CLI, Cloud Provider, Cluster Lifecycle, Instrumentation, Network, Node, Scheduling and Storage] ``
- `- Upgraded CoreDNS to v1.12.4. ([#133968](https://github.com/kubernetes/kubernetes/pull/133968), [@yashsingh74](https://github.com/yashsingh74)) [SIG Cloud Provider and Cluster Lifecycle]`
- `` - Upgraded `CoreDNS` to `v1.12.3`. ([#132288](https://github.com/kubernetes/kubernetes/pull/132288), [@thevilledev](https://github.com/thevilledev)) [SIG Cloud Provider and Cluster Lifecycle] ``
- `` - `etcd: Update etcd to `v3.6.6`. (#135271, @bzsuni) [SIG API Machinery, Cloud Provider, Cluster Lifecycle, Etcd and Testing] ``
- `- cyphar.com/go-pathrs: v0.2.1`
- `- github.com/Masterminds/semver/v3: [v3.4.0](https://github.com/Masterminds/semver/tree/v3.4.0)`
- `- github.com/gkampitakis/ciinfo: [v0.3.2](https://github.com/gkampitakis/ciinfo/tree/v0.3.2)`
- `- github.com/gkampitakis/go-diff: [v1.3.2](https://github.com/gkampitakis/go-diff/tree/v1.3.2)`
- `- github.com/gkampitakis/go-snaps: [v0.5.15](https://github.com/gkampitakis/go-snaps/tree/v0.5.15)`
- `- github.com/goccy/go-yaml: [v1.18.0](https://github.com/goccy/go-yaml/tree/v1.18.0)`
- `- github.com/joshdk/go-junit: [v1.0.0](https://github.com/joshdk/go-junit/tree/v1.0.0)`
- `- github.com/maruel/natural: [v1.1.1](https://github.com/maruel/natural/tree/v1.1.1)`
- `- github.com/mfridman/tparse: [v0.18.0](https://github.com/mfridman/tparse/tree/v0.18.0)`
- `- github.com/moby/sys/atomicwriter: [v0.1.0](https://github.com/moby/sys/tree/atomicwriter/v0.1.0)`
- `- github.com/tidwall/gjson: [v1.18.0](https://github.com/tidwall/gjson/tree/v1.18.0)`
- `- github.com/tidwall/match: [v1.1.1](https://github.com/tidwall/match/tree/v1.1.1)`
- `- github.com/tidwall/pretty: [v1.2.1](https://github.com/tidwall/pretty/tree/v1.2.1)`
- `- github.com/tidwall/sjson: [v1.2.5](https://github.com/tidwall/sjson/tree/v1.2.5)`
- `- go.uber.org/automaxprocs: v1.6.0`
- `- golang.org/x/tools/go/expect: v0.1.1-deprecated`
- `- golang.org/x/tools/go/packages/packagestest: v0.1.1-deprecated`
- `- cloud.google.com/go/compute/metadata: v0.6.0 → v0.7.0`
- `- github.com/aws/aws-sdk-go-v2/config: [v1.27.24 → v1.29.14](https://github.com/aws/aws-sdk-go-v2/compare/config/v1.27.24...config/v1.29.14)`
- `- github.com/aws/aws-sdk-go-v2/credentials: [v1.17.24 → v1.17.67](https://github.com/aws/aws-sdk-go-v2/compare/credentials/v1.17.24...credentials/v1.17.67)`
- `- github.com/aws/aws-sdk-go-v2/feature/ec2/imds: [v1.16.9 → v1.16.30](https://github.com/aws/aws-sdk-go-v2/compare/feature/ec2/imds/v1.16.9...feature/ec2/imds/v1.16.30)`
- `- github.com/aws/aws-sdk-go-v2/internal/configsources: [v1.3.13 → v1.3.34](https://github.com/aws/aws-sdk-go-v2/compare/internal/configsources/v1.3.13...internal/configsources/v1.3.34)`
- `- github.com/aws/aws-sdk-go-v2/internal/endpoints/v2: [v2.6.13 → v2.6.34](https://github.com/aws/aws-sdk-go-v2/compare/internal/endpoints/v2/v2.6.13...internal/endpoints/v2/v2.6.34)`
- `- github.com/aws/aws-sdk-go-v2/internal/ini: [v1.8.0 → v1.8.3](https://github.com/aws/aws-sdk-go-v2/compare/internal/ini/v1.8.0...internal/ini/v1.8.3)`
- `- github.com/aws/aws-sdk-go-v2/service/internal/accept-encoding: [v1.11.3 → v1.12.3](https://github.com/aws/aws-sdk-go-v2/compare/service/internal/accept-encoding/v1.11.3...service/internal/accept-encoding/v1.12.3)`
- `- github.com/aws/aws-sdk-go-v2/service/internal/presigned-url: [v1.11.15 → v1.12.15](https://github.com/aws/aws-sdk-go-v2/compare/service/internal/presigned-url/v1.11.15...service/internal/presigned-url/v1.12.15)`
- `- github.com/aws/aws-sdk-go-v2/service/sso: [v1.22.1 → v1.25.3](https://github.com/aws/aws-sdk-go-v2/compare/service/sso/v1.22.1...service/sso/v1.25.3)`
- `- github.com/aws/aws-sdk-go-v2/service/ssooidc: [v1.26.2 → v1.30.1](https://github.com/aws/aws-sdk-go-v2/compare/service/ssooidc/v1.26.2...service/ssooidc/v1.30.1)`
- `- github.com/aws/aws-sdk-go-v2/service/sts: [v1.30.1 → v1.33.19](https://github.com/aws/aws-sdk-go-v2/compare/service/sts/v1.30.1...service/sts/v1.33.19)`
- `- github.com/aws/aws-sdk-go-v2: [v1.30.1 → v1.36.3](https://github.com/aws/aws-sdk-go-v2/compare/v1.30.1...v1.36.3)`
- `- github.com/aws/smithy-go: [v1.20.3 → v1.22.3](https://github.com/aws/smithy-go/compare/v1.20.3...v1.22.3)`
- `- github.com/containerd/containerd/api: [v1.8.0 → v1.9.0](https://github.com/containerd/containerd/compare/api/v1.8.0...api/v1.9.0)`
- `- github.com/containerd/ttrpc: [v1.2.6 → v1.2.7](https://github.com/containerd/ttrpc/compare/v1.2.6...v1.2.7)`
- `- github.com/containerd/typeurl/v2: [v2.2.2 → v2.2.3](https://github.com/containerd/typeurl/compare/v2.2.2...v2.2.3)`
- `- github.com/coredns/corefile-migration: [v1.0.26 → v1.0.29](https://github.com/coredns/corefile-migration/compare/v1.0.26...v1.0.29)`
- `- github.com/cyphar/filepath-securejoin: [v0.4.1 → v0.6.0](https://github.com/cyphar/filepath-securejoin/compare/v0.4.1...v0.6.0)`
- `- github.com/docker/docker: [v26.1.4+incompatible → v28.2.2+incompatible](https://github.com/docker/docker/compare/v26.1.4...v28.2.2)`
- `- github.com/go-logr/logr: [v1.4.2 → v1.4.3](https://github.com/go-logr/logr/compare/v1.4.2...v1.4.3)`
- `- github.com/google/cadvisor: [v0.52.1 → v0.53.0](https://github.com/google/cadvisor/compare/v0.52.1...v0.53.0)`
- `- github.com/google/pprof: [d1b30fe → 27863c8](https://github.com/google/pprof/compare/d1b30fe...27863c8)`
- `- github.com/onsi/ginkgo/v2: [v2.21.0 → v2.27.2](https://github.com/onsi/ginkgo/compare/v2.21.0...v2.27.2)`
- `- github.com/onsi/gomega: [v1.35.1 → v1.38.2](https://github.com/onsi/gomega/compare/v1.35.1...v1.38.2)`
- `- github.com/opencontainers/cgroups: [v0.0.1 → v0.0.3](https://github.com/opencontainers/cgroups/compare/v0.0.1...v0.0.3)`
- `- github.com/opencontainers/runc: [v1.2.5 → v1.3.0](https://github.com/opencontainers/runc/compare/v1.2.5...v1.3.0)`
- `- github.com/opencontainers/runtime-spec: [v1.2.0 → v1.2.1](https://github.com/opencontainers/runtime-spec/compare/v1.2.0...v1.2.1)`
- `- github.com/opencontainers/selinux: [v1.11.1 → v1.13.0](https://github.com/opencontainers/selinux/compare/v1.11.1...v1.13.0)`
- `- github.com/prometheus/client_golang: [v1.22.0 → v1.23.2](https://github.com/prometheus/client_golang/compare/v1.22.0...v1.23.2)`
- `- github.com/prometheus/client_model: [v0.6.1 → v0.6.2](https://github.com/prometheus/client_model/compare/v0.6.1...v0.6.2)`
- `- github.com/prometheus/common: [v0.62.0 → v0.66.1](https://github.com/prometheus/common/compare/v0.62.0...v0.66.1)`
- `- github.com/prometheus/procfs: [v0.15.1 → v0.16.1](https://github.com/prometheus/procfs/compare/v0.15.1...v0.16.1)`
- `- github.com/rogpeppe/go-internal: [v1.13.1 → v1.14.1](https://github.com/rogpeppe/go-internal/compare/v1.13.1...v1.14.1)`
- `- github.com/spf13/cobra: [v1.9.1 → v1.10.0](https://github.com/spf13/cobra/compare/v1.9.1...v1.10.0)`
- `- github.com/spf13/pflag: [v1.0.6 → v1.0.9](https://github.com/spf13/pflag/compare/v1.0.6...v1.0.9)`
- `- github.com/stretchr/testify: [v1.10.0 → v1.11.1](https://github.com/stretchr/testify/compare/v1.10.0...v1.11.1)`
- `- go.etcd.io/bbolt: v1.4.2 → v1.4.3`
- `- go.etcd.io/etcd/api/v3: v3.6.4 → v3.6.5`
- `- go.etcd.io/etcd/client/pkg/v3: v3.6.4 → v3.6.5`
- `- go.etcd.io/etcd/client/v3: v3.6.4 → v3.6.5`
- `- go.etcd.io/etcd/pkg/v3: v3.6.4 → v3.6.5`
- `- go.etcd.io/etcd/server/v3: v3.6.4 → v3.6.5`
- `- go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp: v0.58.0 → v0.61.0`
- `- go.opentelemetry.io/otel/metric: v1.35.0 → v1.36.0`
- `- go.opentelemetry.io/otel/sdk/metric: v1.34.0 → v1.36.0`
- `- go.opentelemetry.io/otel/sdk: v1.34.0 → v1.36.0`
- `- go.opentelemetry.io/otel/trace: v1.35.0 → v1.36.0`
- `- go.opentelemetry.io/otel: v1.35.0 → v1.36.0`
- `- go.yaml.in/yaml/v2: v2.4.2 → v2.4.3`
- `- golang.org/x/crypto: v0.36.0 → v0.45.0`
- `- golang.org/x/mod: v0.21.0 → v0.29.0`
- `- golang.org/x/net: v0.38.0 → v0.47.0`
- `- golang.org/x/oauth2: v0.27.0 → v0.30.0`
- `- golang.org/x/sync: v0.12.0 → v0.18.0`
- `- golang.org/x/sys: v0.31.0 → v0.38.0`
- `- golang.org/x/telemetry: bda5523 → 078029d`
- `- golang.org/x/term: v0.30.0 → v0.37.0`
- `- golang.org/x/text: v0.23.0 → v0.31.0`
- `- golang.org/x/tools: v0.26.0 → v0.38.0`
- `- google.golang.org/genproto/googleapis/rpc: a0af3ef → 200df99`
- `- google.golang.org/grpc: v1.72.1 → v1.72.2`
- `- google.golang.org/protobuf: v1.36.5 → v1.36.8`
- `- gopkg.in/evanphx/json-patch.v4: v4.12.0 → v4.13.0`
- `- k8s.io/gengo/v2: 85fd79d → ec3ebc5`
- `- k8s.io/kube-openapi: f3f2b99 → 589584f`
- `- k8s.io/system-validators: v1.10.1 → v1.12.1`
- `- k8s.io/utils: 4c0f3b2 → bc988d5`
- `- sigs.k8s.io/json: cfa47c3 → 2d32026`
- `### Removed - gopkg.in/yaml.v2: v2.4.0`

## Other changes (74)

- Scheduler scoring for prioritized-list requests
- kube-scheduler enables `NominatedNodeNameForExpectation` by default at beta
- `ClearingNominatedNodeNameAfterBinding` promoted to beta and enabled by default in kube-apiserver
- Custom uniqueness validation control for listmap fields
- kube-apiserver CRD format warnings fixed
- Kubelet image-pull tracking moves to `v1beta1`
- ReplicaSet and Deployment terminating replica tracking promoted to beta and enabled by default
- StatefulSet rolling-update `maxUnavailable` promoted to beta and enabled by default
- `KubeletEnsureSecretPulledImages` promoted to beta and enabled by default
- `MutatingAdmissionPolicy` storage version changed to `v1beta1`
- Stale alpha description removed for `PodObservedGenerationTracking`
- kubelet stress test for Pod cleanup after `VolumeAttachmentLimitExceeded` rejection
- Additional events for Pod resize status transitions
- Scheduler `statusz` lists registered paths
- Scheduler `SchedulerAsyncAPICalls` gate re-enabled by default
- Kubernetes controllers now use `WaitForNamedCacheSyncWithContext`
- Beta `MutableCSINodeAllocatableCount` feature gate enabled by default
- `WatchListClient` enabled by default
- `ContainerRestartRules` probe behavior fixed
- `PodTopologyLabelsAdmission` promoted to beta and enabled by default
- Image volume sources promoted to beta and enabled by default
- Scheduler gains opportunistic batching
- Higher-throughput `real-FIFO` queue processing
- End-to-end tests for component invariant metrics added
- The `kubelet-wait-bootstrap` subphase now always runs during `kubeadm join`
- Kubernetes build updated to Go 1.25.4
- `KubeletCrashLoopBackOffMax` promoted to beta and enabled by default
- Deterministic device class selection for extended resources
- Scheduler clears Pod `nominatedNodeName` after scheduling or binding failure
- `applyconfiguration-gen` preserves source comments in generated code
- kubeadm adds a `RuntimeConfig` support preflight check
- kubectl `--chunk-size` flag promoted to stable
- More specific errors for failSafe-blocked evictions
- DRA requests drop `tolerations` when `DRADeviceTaints` is disabled
- kube-scheduler preemptor retry behavior restored
- Duplicate validation removed from `PersistentVolumeClaims`, `VolumeAttachments`, and `VolumeAttributesClasses` updates
- Duplicate validation on `Role` and `RoleBinding` updates removed
- API Dispatcher binding throughput degradation addressed
- Metrics registration data race fixed
- ResourceClaim strategy update errors fixed with `DRAConsumableCapacity` disabled
- Scheduler `VolumeBinding` cache scheduling delay fixed
- API replacement tags no longer exclude versions from storage-version selection
- client-go/cli-runtime certificate and key configuration merge validation fixed
- Asynchronous pod-to-service lookup in `Endpoint` and `EndpointSlice` controllers
- More actionable `FreeDiskSpaceFailed` event details
- Spurious kube-apiserver key-error logs fixed
- kube-apiserver reports failed values in CEL validation errors
- Kube-apiserver alpha API startup warnings removed
- Garbage collection node-printing data race fixed
- Kubeadm IPv6 hostname construction failure fixed
- `VolumeAttributesClass` API lifecycle test promoted to conformance
- Default `SessionAffinity` warning removed for headless Services
- Typed workqueue shutdown waits for managed goroutines
- Volume e2e framework adds a `Step` field for expansion increments
- `kubeadm` static pod manifest comparison now uses FNV-1a
- Endpoint and `EndpointSlice` hashes switched from MD5 to FNV-1a
- MD5 lint rule added and storage e2e integrity test switched to SHA-256
- `nfacct` test cases fixed on s390x
- Go API deprecation formatting fixed and linting enabled
- HPA container-specific resource metric lookup now exits early
- HPA calculation reports missing containers
- Test coverage increased to 89.8%
- Kubeadm raises supported etcd version to `v3.5.23`
- Kubelet `CPUManager` logs migrated to contextual logging
- CPU Manager `strict-cpu-reservation` option reaches GA
- Topology Manager `max-allowable-numa-nodes` option reaches GA
- Reduced detail in Portworx volume-operation error events
- Kubernetes builds no longer depend on `rsync`
- Metrics documentation now includes deprecated `apiserver_storage_objects` metric
- Kubernetes build process runs pre-built images directly without `rsyncd`
- Tests switched to Go 1.25 container-aware GOMAXPROCS behavior
- `kubectl auth reconcile` retries on conflicts
- `kubectl wait` description no longer marked Experimental
- `kubectl wait` short description no longer says Experimental

---

Summaries are produced by AI, so they can be wrong, and they are provided without warranty.
