- Fixed the following 
  -  AWS Cloud Provider attempts to delete LoadBalancer security group it didn’t provision

from './methods/enableEventListeners';\nimport disableEventListeners from './methods/disableEventListeners';\nimport Defaults from './methods/defaults';\nimport placements from './methods/placements';\n\nexport default class Popper {\n  /**\n   * Creates a new Popper.js instance.\n   * @class Popper\n   * @param {Element|referenceObject} reference - The reference element used to position the popper\n   * @param {Element} popper - The HTML / XML element used as the popper\n   * @param {Object} options...

* Add option to disable persistence storage for etcd ([#40862](https://github.com/kubernetes/kubernetes/pull/40862), [@shashidharatd](https://github.com/shashidharatd))

    inputStream.close()
    val elapsedNanos = System.nanoTime() - startNanos
    val elapsedMillis = TimeUnit.NANOSECONDS.toMillis(elapsedNanos)

    // If we're working correctly, this should be greater than 100ms, but less than double that.
    // Previously we had a bug where we would download the entire response body as long as no
    // individual read took longer than 100ms.
    assertThat(elapsedMillis).isLessThan(500L)

- For CSI drivers, kubelet no longer creates the target_path for NodePublishVolume in accordance with the CSI spec. Kubelet also no longer checks if staging and target paths are mounts or corrupted. CSI drivers need to be idempotent and do any necessary mount verification. ([#88759](https:/...

- **Federation**
  - [alpha] Creating a `Federated Ingress` is as simple as submitting
    an `Ingress` creation request to the Federation API Server. The
    Federation control system then creates and maintains a single
    global virtual IP to load balance incoming HTTP(S) traffic across
    some or all the registered clusters, across all regions. Google's
    GCE L7 LoadBalancer is the first supported implementation, and

* Fix issue in reconstruct volume data when kubelet restarts ([#36616](https://github.com/kubernetes/kubernetes/pull/36616), [@jingxu97](https://github.com/jingxu97))
* Ensure proper serialization of updates and creates in federation test watcher ([#36613](https://github.com/kubernetes/kubernetes/pull/36613), [@mwielgus](https://github.com/mwielgus))

    *   Unary GPU kernels: Abs, Atanh, Acos, Acosh, Asin, Asinh, Atan, Cos, Cosh, Sin, Sinh, Tan, Tanh.
    *   Binary GPU kernels: AddV2, Sub, Div, DivNoNan, Mul, MulNoNan, FloorDiv, Equal, NotEqual, Greater, GreaterEqual, LessEqual, Less.

* `tf.lite`
    * Add experimental supports conversion of models that may be larger than 2GB before buffer deduplication

### Bug Fixes and Other Changes

- If the parent directory of the file specified in the `--audit-log-path` argument does not exist, Kubernetes now creates it. ([#111226](https://github.com/kubernetes/kubernetes/pull/111226), [@vpnachev](https://github.com/vpnachev)) [SIG Auth]

- Update kube-apiserver's priority & fairness work estimator such that 'max seats' is MIN(0.15 x nominalCL, nominalCL / handSize)
  
  This fixes a bug where clients with requests using hand size x max seats greater than the nominal concurrency limit can starve other requests in the same priority level. ([#118601](https://github.com/kubernetes/kubernetes/pull/118601), [@andrewsykim](https://github.com/andrewsykim)) [SIG API Machinery]