* Splits Juju Charm layers into master/worker roles ([#40324](https://github.com/kubernetes/kubernetes/pull/40324), [@chuckbutler](https://github.com/chuckbutler))
    * Adds support for 1.5.x series of Kubernetes
    * Introduces a tactic for keeping templates in sync with upstream eliminating template drift
    * Adds CNI support to the Juju Charms
    * Adds durable storage support to the Juju Charms

* kubectl: improves compatibility with older servers when creating/updating API objects ([#61949](https://github.com/kubernetes/kubernetes/pull/61949), [@liggitt](https://github.com/liggitt))
* Bound cloud allocator to 10 retries with 100 ms delay between retries. ([#61375](https://github.com/kubernetes/kubernetes/pull/61375), [@satyasm](https://github.com/satyasm))

* Fix a bug that prevents client-go metrics from being registered in prometheus in multiple components. ([#53434](https://github.com/kubernetes/kubernetes/pull/53434), [@crassirostris](https://github.com/crassirostris))
* Webhook always retries connection reset error. ([#53947](https://github.com/kubernetes/kubernetes/pull/53947), [@crassirostris](https://github.com/crassirostris))

  - Serve from cache for the whole period of ongoing throttling
  - Fix unwanted VMSS VMs cache invalidations
  - Enforce setting the number of retries if cloud provider backoff is enabled
  - Don't update capacity if VMSS provisioning state is updating
  - Support allocatable resources overrides via VMSS tags
  - Add missing stable labels in template nodes

- Ubuntu 18.04 (Bionic) series has been added to Juju charms ([#65644](https://github.com/kubernetes/kubernetes/pull/65644), [@tvansteenburgh](https://github.com/tvansteenburgh))

  - If all QueueingHintFn return QueueSkip, it puts this pod back to the unschedulable pod pool
  
  Having appropriate QueueingHintFn contributes to reducing useless retries and thus improves the overall scheduler's performance.
  
  **How can I migrate?**
  
  For backward compatibility, nil `QueueingHintFn` is treated as always returning QueueAfterBackoff. 

- Promoting `WindowsHostProcessContainers` to stable. ([#113476](https://github.com/kubernetes/kubernetes/pull/113476), [@marosset](https://github.com/marosset))
- Scheduler now retries updating a pod's status on `ServiceUnavailable` and `InternalError` errors, in addition to `net.ConnectionRefused` error. ([#111809](https://github.com/kubernetes/kubernetes/pull/111809), [@Huang-Wei](https://github.com/Huang-Wei))

        * `{WorkQueueName}_retries`
    * Please convert to the following metrics:
        * `workqueue_depth`
        * `workqueue_adds_total`
        * `workqueue_queue_latency_seconds`
        * `workqueue_work_duration_seconds`
        * `workqueue_unfinished_work_seconds`
        * `workqueue_longest_running_processor_seconds`
        * `workqueue_retries_total`

- `scheduler`: handling of unschedulable pods because a `ResourceClass` is missing
  is a bit more efficient and no longer relies on periodic retries. ([#120213](https://github.com/kubernetes/kubernetes/pull/120213), [@pohly](https://github.com/pohly))

## Dependencies

### Added
- cloud.google.com/go/dataproc/v2: v2.0.1