- Don't update capacity if VMSS provisioning state is updating
  - Support allocatable resources overrides via VMSS tags
  - Add missing stable labels in template nodes
  - Proactively set instance status to deleting on node deletions
  
  ### Cluster API
  - Migrate interaction with the API from using internal types to using Unstructured
  - Improve tests to work better with constrained resources
  - Add support for node autodiscovery

- Fix JobTrackingWithFinalizers that:
  - was declaring a job finished before counting all the created pods in the status
  - was leaving pods with finalizers, blocking pod and job deletions
  
  JobTrackingWithFinalizers is still disabled by default. ([#109486](https://github.com/kubernetes/kubernetes/pull/109486), [@alculquicondor](https://github.com/alculquicondor)) [SIG Apps and Testing]

import   "lib/math"         math.Sin
import m "lib/math"         m.Sin
import . "lib/math"         Sin
</pre>

<p>
An import declaration declares a dependency relation between
the importing and imported package.
It is illegal for a package to import itself, directly or indirectly,
or to directly import a package without

- Fix network_programming_latency metric reporting for Endpoints/EndpointSlice deletions, where we don't have correct timestamp ([#95363](https://github.com/kubernetes/kubernetes/pull/95363), [@wojtek-t](https://github.com/wojtek-t)) [SIG Network and Scalability]

### Bug or Regression

- Fix JobTrackingWithFinalizers that:
  - was declaring a job finished before counting all the created pods in the status
  - was leaving pods with finalizers, blocking pod and job deletions
  
  JobTrackingWithFinalizers is still disabled by default. ([#109486](https://github.com/kubernetes/kubernetes/pull/109486), [@alculquicondor](https://github.com/alculquicondor)) [SIG Apps and Testing]

    *   Modify custom export strategy to account for multidimensional sparse
        float splits.
    *   `Conv2D`, `Conv2DBackpropInput`, `Conv2DBackpropFilter` now supports
        arbitrary dilations with GPU and cuDNNv6 support.
    *   `Estimator` now supports `Dataset`: `input_fn` can return a `Dataset`
        instead of `Tensor`s.
    *   Add `RevBlock`, a memory-efficient implementation of reversible residual