- Kube-controller-manager: the `horizontal-pod-autoscaler-upscale-delay`...

<a href="https://en.wikipedia.org/wiki/UTF-8">UTF-8</a>. The text is not
canonicalized, so a single accented code point is distinct from the
same character constructed from combining an accent and a letter;
those are treated as two code points.  For simplicity, this document
will use the unqualified term <i>character</i> to refer to a Unicode code point
in the source text.
</p>
<p>

- Fixed attaching volumes after detach errors. Now volumes that failed to detach are not treated as attached, Kubernetes will make sure they are fully attached before they can be used by pods. ([#120595](https://github.com/kubernetes/kubernetes/pull/120595), [@jsafrane](https://github.com/jsafrane)) [SIG Apps and Storage]

- The `storageVersionHash` feature is now beta. `StorageVersionHash` is a field in the discovery document of each resource. It enables clients to detect whether the storage version of that resource has changed. Its value must be treated as opaque by clients. Only equality comparison on the value is valid. ([#78325](https://github.com/kubernetes/kubernetes/pull/78325), [@caesarxuchao](https://github.com/caesarxuchao))

- Kube-apiserver: resolved a regression that treated `304 Not Modified` responses from aggregated API servers as internal errors ([#112530](https://github.com/kubernetes/kubernetes/pull/112530), [@liggitt](https://github.com/liggitt)) [SIG API Machinery]

- Fixed attaching volumes after detach errors. Now volumes that failed to detach are not treated as attached, Kubernetes will make sure they are fully attached before they can be used by pods. ([#120595](https://github.com/kubernetes/kubernetes/pull/120595), [@jsafrane](https://github.com/jsafrane)) [SIG Apps and Storage]

- Fixed attaching volumes after detach errors. Now volumes that failed to detach are not treated as attached. Kubernetes will make sure they are fully attached before they can be used by pods. ([#120595](https://github.com/kubernetes/kubernetes/pull/120595), [@jsafrane](https://github.com/jsafrane))

- 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))

        structures of `tensors` as inputs (instead of just a list of tensors).
        Note that Python structures such as tuples and lists now won't be
        treated as tensors, so if you still want them to be treated that way,
        you need to wrap them with `tf.convert_to_tensor`.
    *   No lowering on gradient case op when input is `DeviceIndex` op.

- The terminationGracePeriodSeconds...