* apiserver: fixes handling and logging of panics in REST handlers to prevent crashes ([#71076](https://github.com/kubernetes/kubernetes/pull/71076), [@liggitt](https://github.com/liggitt))

* apiserver: fixes handling and logging of panics in REST handlers to prevent crashes ([#71076](https://github.com/kubernetes/kubernetes/pull/71076), [@liggitt](https://github.com/liggitt))

* apiserver: fixes handling and logging of panics in REST handlers to prevent crashes ([#71076](https://github.com/kubernetes/kubernetes/pull/71076), [@liggitt](https://github.com/liggitt))

* Restores the ability of older clients to delete and scale jobs with initContainers ([#59880](https://github.com/kubernetes/kubernetes/pull/59880), [@liggitt](https://github.com/liggitt))
* Fix race causing apiserver crashes during etcd healthchecking ([#60069](https://github.com/kubernetes/kubernetes/pull/60069), [@wojtek-t](https://github.com/wojtek-t))

- In some cases, a Flex volume resize may leave a PVC with erroneous Resizing condition even after volume has been successfully expanded. Users may choose to delete the condition, but it is not required. ([#71470](https://github.com/kubernetes/kubernetes/issues/71470))

- Adopted a new implementation of watch caches for **list** verbs, using a btree data structure. The new implementation is active by default; you can opt out by disabling the `BtreeWatchCache` feature gate. ([#128415](https://github.com/kubernetes/kubernetes/pull/128415), [@serathius](https:/...

      Post this PR, when users supply 0 for CacheSize this will result in a validation error.
   2. CacheSize type was changed from int32 to *int32. This allows defaulting logic to differentiate between cases where users explicitly supplied 0 vs. not supplied any value.

- Fix a race in the timeout handler that could lead to kube-apiserver crashes ([#108455](https://github.com/kubernetes/kubernetes/pull/108455), [@Argh4k](https://github.com/Argh4k))

  
      `kube-proxy --feature-gates NFTablesProxyMode=true --proxy-mode nftables`
  
  This is currently an alpha-level feature and while it probably will not
  eat your data, it may nibble at it a bit. (It passes e2e testing but has
  not yet seen real-world use.)
  
  At this point it should be functionally mostly identical to the iptables
  mode, except that it does not (and will not) support Service NodePorts on

    *  - vSphere cloud provider (VCP) uses OS hostname and not vSphere Inventory VM Name.
    *    That means, now  VCP can handle cases where user changes VM inventory name.
    * - VCP can handle cases where VM migrates to other ESXi cluster or datacenter or vCenter.
    * The only requirement is the shared storage. VCP needs shared storage on all Node VMs.