* Reduce Metrics Server memory requirement from 140Mi + 4Mi per node to 40Mi + 4Mi per node. ([#58391](https://github.com/kubernetes/kubernetes/pull/58391), [@kawych](https://github.com/kawych))

- The `apiserver` exposes 4 new metrics that allow to track the status of the Service CIDRs allocations:
      - current number of available IPs per Service CIDR
      - current number of used IPs per Service CIDR
      - total number of allocation per Service CIDR
      - total number of allocation errors per ServiceCIDR ([#104119](https://github.com/kubernetes/kubernetes/pull/104119), [@aojea](https://github.com/aojea))

* Add MAX_PODS_PER_NODE env so that GCE/GKE user can use it to specify the default max pods per node for the cluster. IP_ALIAS_SIZE will be changed accordingly. Must have ip alias enabled. ([#63451](https://github.com/kubernetes/kubernetes/pull/63451), [@grayluck](https://github.com/grayluck))

- Kube-scheduler: in Kubernetes 1.33, the number of devices that can be allocated per ResourceClaim was accidentally reduced to 16. Now the supported number of devices per ResourceClaim is 32 again. ([#131662](https://github.com/kubernetes/kubernetes/pull/131662), [@mortent](https://github.com/mortent)) [SIG Node]

- Kube-scheduler: in Kubernetes 1.33, the number of devices that can be allocated per ResourceClaim was accidentally reduced to 16. Now the supported number of devices per ResourceClaim is 32 again. ([#131679](https://github.com/kubernetes/kubernetes/pull/131679), [@mortent](https://github.com/mortent)) [SIG Node]

	type parameter, and all the types in its type set unify with
	the other type
	per the unification rules for <code>≡<sub>A</sub></code>
	(loose unification at the top level and exact unification
	for element types).
</li>
</ul>

<p>
If both types are bound type parameters, they unify per the given
matching modes if:
</p>

<ul>
<li>
	Both type parameters are identical.

## Changelog since v1.8.12

### Other notable changes

* Add MAX_PODS_PER_NODE env so that user can use it to specify the default max pods per node for the cluster. IP_ALIAS_SIZE will be changed accordingly. Must have ip alias enabled. ([#63466](https://github.com/kubernetes/kubernetes/pull/63466), [@grayluck](https://github.com/grayluck))

### Cloud Provider
- Azure: update disk lock logic per vm during attach/detach to allow concurrent updates for different nodes. ([#85115](https://github.com/kubernetes/kubernetes/pull/85115), [@aramase](https://github.com/aramase))

  - Clients should reload the token from disk periodically (once per minute is recommended) to ensure they continue to use a valid token. `k8s.io/client-go` version v11.0.0+ and v0.15.0+ reload tokens automatically.

- Fixed an issue in the Windows kube-proxy (winkernel) where IPv4 and IPv6 Service load balancers could be incorrectly shared, causing broken dual-stack Service behavior. The kube-proxy now tracks load balancers per IP family, enabling correct support for PreferDualStack and RequireDualStack Services on Windows nodes. ([#136373](https://github.com/kubernetes/kubernetes/pull/136373), [@princepereira](https://github.com/princepereira)) [SIG Network and Windows]