- Added validation to reject Pods using the `PodLevelResources` feature on Windows OS due to lack of support. The API server rejected Pods with pod-level resources and a `Pod.spec.os.name` targeting Windows. Kubelet on nodes running Windows also rejected Pods with pod-level resources at the admission phase. ([#133046](https://github.com/kubernetes/kubernetes/pull/133046), [@toVersus](https://github.com/toVersus)) [SIG Apps and Node]

- 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))

### Bug or Regression

## Changelog since v1.13.10

### Other notable changes

* Fix a bug in apiserver that could cause a valid update request to be rejected with a precondition check failure. ([#82303](https://github.com/kubernetes/kubernetes/pull/82303), [@roycaihw](https://github.com/roycaihw))

- Kube-apiserver: Get, GetList and Watch requests that should be served by the apiserver cacher during shutdown will be rejected to avoid a deadlock situation leaving requests hanging. ([#108414](https://github.com/kubernetes/kubernetes/pull/108414), [@aojea](https://github.com/aojea))

t.Fatal(err) } else if k.IsZero() == 1 { t.Error("k is zero") } else if p.Bytes()[0] != 4 { t.Error("p is infinity") } if loopCount == 0 { t.Error("overflow was not rejected") } loopCount = 0 // A sequence of all zeroes will generate zero, which should be rejected. r = io.MultiReader(bytes.NewReader(bytes.Repeat([]byte{0}, 100)), rand.Reader) if k, p, err := randomPoint(c, func(b []byte) error { _, err := r.Read(b) return err }); err != nil { t.Fatal(err) } else if k.IsZero() == 1 { t.Error("k is zero")...

* Fix a bug in apiserver that could cause a valid update request to be rejected with a precondition check failure. ([#82303](https://github.com/kubernetes/kubernetes/pull/82303), [@roycaihw](https://github.com/roycaihw))

- The scheduling queue didn't notice any extenders' failures, it could miss some cluster events,
  and it could end up Pods rejected by Extenders stuck in unschedulable pod pool in 5min in the worst-case scenario.
  Now, the scheduling queue notices extenders' failures and requeue Pods rejected by Extenders appropriately. ([#122043](https://github.com/kubernetes/kubernetes/pull/122043), [@sanposhiho](https://github.com/sanposhiho)) [SIG Scheduling]

* Fix bug with service nodeports that have no backends not being rejected, when they should be.  This is not a regression vs v1.5 - it's a fix that didn't quite fix hard enough. ([#43972](https://github.com/kubernetes/kubernetes/pull/43972), [@thockin](https://github.com/thockin))

* Fixes an initializer bug where update requests which had an empty pending initializers list were erroneously rejected. ([#52558](https://github.com/kubernetes/kubernetes/pull/52558), [@jennybuckley](https://github.com/jennybuckley))