* Till this release, iscsi plugin was waiting 10 seconds for a path to appear in the device list. However this timeout is not enough or less than default device discovery timeout in most of the systems which cause certain device to be not accounted for the volume. This timeout has been lifted to 30seconds from this release and it should help to avoid mount issues due to device discovery. ([#78475](https://github.com/kubernetes/kubernetes/pull/78475), [@humblec](https://github.com/humblec))...

  TFE_Context* ctx_1 =
      CreateContext(serialized_server_def_1,
                    /*isolate_session_state=*/false, init_timeout_in_ms);

  // Remote device on `worker2`.
  const char remote_device[] = "/job:localhost/replica:0/task:2/device:CPU:0";
  // `ctx_0`, `ctx_1` contains `remote_device`.
  {
    const std::vector<std::string>& device_names = ListDeviceNames(ctx_0);

        as arg with default value `None` to choose the I/O device for loading
        models and weights.
    *   Update `tf.saved_model.SaveOptions` with
        [`experimental_io_device`](https://www.tensorflow.org/versions/r2.3/api_docs/python/tf/saved_model/SaveOptions?hl=en)
        as arg with default value `None` to choose the I/O device for saving
        models and weights.

* Fix issue where on re-registration of device plugin, `allocatable` was not getting updated. This issue makes devices invisible to the Kubelet if device plugin restarts. Only work-around, if this fix is not there, is to restart the kubelet and then start device plugin. ([#63118](https://github.com/kubernetes/kubernetes/pull/63118), [@vikaschoudhary16](https://github.com/vikaschoudhary16))

* Backporting TPR fix to 1.4 ([#42380](https://github.com/kubernetes/kubernetes/pull/42380), [@foxish](https://github.com/foxish))
* Fix AWS device allocator to only use valid device names ([#41455](https://github.com/kubernetes/kubernetes/pull/41455), [@gnufied](https://github.com/gnufied))

  // ratio between the target value and the current value by the current
  // number of pods.  Ergo, metrics used must decrease as the pod count is
  // increased, and vice-versa.  See the individual metric source types for
  // more information about how each type of metric must respond.
  // If not set, the default metric will be set to 80% average CPU utilization.
  // +optional

				data, origdata)
		}
	}
	if entriesLeft > 0 {
		t.Fatalf("not all entries are in the archive")
	}
}

func TestWriterAddFSNonRegularFiles(t *testing.T) {
	fsys := fstest.MapFS{
		"device":  {Data: []byte("hello"), Mode: 0755 | fs.ModeDevice},
		"symlink": {Data: []byte("world"), Mode: 0755 | fs.ModeSymlink},
	}
	var buf bytes.Buffer
	tw := NewWriter(&buf)
	if err := tw.AddFS(fsys); err == nil {

## Changelog since v1.3.7

### Other notable changes

* AWS: fix volume device assignment race condition ([#31090](https://github.com/kubernetes/kubernetes/pull/31090), [@justinsb](https://github.com/justinsb))



# v1.3.7

* Fix device mountable volume names in DSW to prevent races in device mountable plugin, e.g. local. ([#71509](https://github.com/kubernetes/kubernetes/pull/71509), [@cofyc](https://github.com/cofyc))

  /*
   * We have some considerable complexity in these create methods because of
   * Builder.buildKeepingLast(). The same Builder might be called with buildKeepingLast() and then
   * buildOrThrow(), or vice versa. So in particular, if we modify alternatingKeysAndValues to
   * eliminate duplicate keys (for buildKeepingLast()) then we have to ensure that a later call to