## Changelog since v1.12.0-beta.2

### Action Required

* Service events are now added in azure-cloud-provider for easily identify the underground errors of Azure API. ([#68212](https://github.com/kubernetes/kubernetes/pull/68212), [@feiskyer](https://github.com/feiskyer))
    * Action required: The following clusterrole and clusterrolebinding should be applied:
    ```
         kind: List
         apiVersion: v1

### Cluster Load Balancing
* Update defaultbackend image to 1.3 ([#42212](https://github.com/kubernetes/kubernetes/pull/42212), [@timstclair](https://github.com/timstclair))

### etcd Empty Dir Cleanup

* fix WaitForAttach failure issue for azure disk ([#62612](https://github.com/kubernetes/kubernetes/pull/62612), [@andyzhangx](https://github.com/andyzhangx))

71 e9 64 da dc 92 19 ed 30 e1 7f 06 b4 b6 8a 95 5c 16 b3 12 b1 ed df 02 8b 74 97 6b ed 6b 3f 6a 63 d4 e7 78 59 24 3c 9c cc dc 98 01 65 23 ab b0 24 83 b3 55 91 c3 3a ad 81 21 3b b7 c7 bb 1a 47 0a ab c1 0d 44 25 6c 4d 45 59 d9 16 # Salt: ef a8 bf f9 62 12 b2 f4 a3 f3 71 a1 0d 57 41 52 65 5f 5d fb # Signature: 7e 09 35 ea 18 f4 d6 c1 d1 7c e8 2e b2 b3 83 6c 55 b3 84 58 9c e1 9d fe 74 33 63 ac 99 48 d1 f3 46 b7 bf dd fe 92 ef d7 8a db 21 fa ef c8 9a de 42 b1 0f 37 40 03 fe 12 2e 67 42 9a 1c b8 cb d1 f8...

Effect in increasing the Length of the coloured Spectrum. For I tried
the same Experiment with other Prisms with the same Success. And
particularly with a Prism which seemed free from such Veins, and whose
refracting Angle was 62-1/2 Degrees, I found the Length of the Image
9-3/4 or 10 Inches at the distance of 18-1/2 Feet from the Prism, the
Breadth of the Hole in the Window-shut being 1/4 of an Inch, as before.

  // CHECK: mhlo.convolution(%arg0, %arg1)
  // CHECK-SAME:  batch_group_count = 2 : i64
  // CHECK-SAME:  feature_group_count = 1 : i64

  %filter_sizes = "tf.Const" () { value = dense<[2, 2, 1, 2]> : tensor<4xi32> } : () -> tensor<4xi32>
  %result = "tf.Conv2DBackpropFilter"(%input, %filter_sizes, %out_backprop) {
    data_format = "NHWC",
    dilations = [1, 1, 1, 1],
    explicit_paddings = [],

  %21 = "tf.ExpandDims"(%20, %7) {device = ""} : (tensor<1xi32>, tensor<i32>) -> tensor<1x1xi32>
  %22 = "tf.Cast"(%21) {Truncate = false, device = ""} : (tensor<1x1xi32>) -> tensor<1x1xi64>
  %23 = "tf.Reshape"(%22, %12) {device = ""} : (tensor<1x1xi64>, tensor<1xi64>) -> tensor<1xi64>
  %24 = "tf.Reshape"(%19, %5) {device = ""} : (tensor<1x!tf_type.string>, tensor<1xi32>) -> tensor<1x!tf_type.string>