- Fix: use force detach for azure disk ([#91948](https://github.com/kubernetes/kubernetes/pull/91948), [@andyzhangx](https://github.com/andyzhangx)) [SIG Cloud Provider]
- Fixes a problem with 63-second or 1-second connection delays with some VXLAN-based
  network plugins which was first widely noticed in 1.16 (though some users saw it
  earlier than that, possibly only with specific network plugins). If you were previously

- Fixed missing `kubelet_volume_stats_*` metrics. ([#133890](https://github.com/kubernetes/kubernetes/pull/133890), [@huww98](https://github.com/huww98)) [SIG Instrumentation and Node]
- Fixed occasional schedule delays when a static `PersistentVolume` is created. ([#133929](https://github.com/kubernetes/kubernetes/pull/133929), [@huww98](https://github.com/huww98)) [SIG Scheduling and Storage]

- Increased the maximum backoff delay of the endpointslice controller to match the expected sequence of delays when syncing Services. ([#112353](https://github.com/kubernetes/kubernetes/pull/112353), [@dgrisonnet](https://github.com/dgrisonnet))

* Set route_localnet on nodes & masters in GCE ([#55004](https://github.com/kubernetes/kubernetes/pull/55004), [@ihmccreery](https://github.com/ihmccreery))
* Configurable liveness probe initial delays for etcd and kube-apiserver in GCE ([#57749](https://github.com/kubernetes/kubernetes/pull/57749), [@wojtek-t](https://github.com/wojtek-t))

- Failed pods associated with a job with `parallelism = 1` are recreated by the job controller honoring exponential backoff delay again. However, for jobs with `parallelism > 1`, pods might be created without exponential backoff delay. ([#115021](https://github.com/kubernetes/kubernetes/pull/115021), [@nikhita](https://github.com/nikhita)) [SIG Apps]

### Bug or Regression

- Fix bug that prevents the alpha feature PodTopologyLabelAdmission to not work due to checking for the incorrect label key when copying topology labels. This bug delays the graduation of the feature to beta by an additional release to allow time for meaningful feedback. ([#132462](https://github.com/kubernetes/kubernetes/pull/132462), [@munnerz](https://github.com/munnerz)) [SIG Node]

* GCE: Fixes ILB creation on automatic networks with manually created subnetworks. ([#57351](https://github.com/kubernetes/kubernetes/pull/57351), [@nicksardo](https://github.com/nicksardo))
* Configurable liveness probe initial delays for etcd and kube-apiserver in GCE ([#57749](https://github.com/kubernetes/kubernetes/pull/57749), [@wojtek-t](https://github.com/wojtek-t))

* Free up CPU and memory requested but unused by Metrics Server Pod Nanny. ([#57252](https://github.com/kubernetes/kubernetes/pull/57252), [@kawych](https://github.com/kawych))
* Configurable liveness probe initial delays for etcd and kube-apiserver in GCE ([#57749](https://github.com/kubernetes/kubernetes/pull/57749), [@wojtek-t](https://github.com/wojtek-t))

- Fixes a problem with 63-second or 1-second connection delays with some VXLAN-based
  network plugins which was first widely noticed in 1.16 (though some users saw it
  earlier than that, possibly only with specific network plugins). If you were previously

- Failed pods associated with a job with `parallelism = 1` are recreated by the job controller honoring exponential backoff delay again. However, for jobs with `parallelism > 1`, pods might be created without exponential backoff delay. ([#115020](https://github.com/kubernetes/kubernetes/pull/115020), [@nikhita](https://github.com/nikhita)) [SIG Apps]