* Self hosted is no longer supported in the standard workflow. The feature flags have been removed and your self hosted cluster is no longer able to upgrade via kubeadm. ([#69878](https://github.com/kubernetes/kubernetes/pull/69878), [@Klaven](https://github.com/Klaven))

 *
 * <p>Subclasses are guaranteed that the life cycle methods ({@link #runOneIteration}, {@link
 * #startUp} and {@link #shutDown}) will never run concurrently. Notably, if any execution of {@link
 * #runOneIteration} takes longer than its schedule defines, then subsequent executions may start
 * late. Also, all life cycle methods are executed with a lock held, so subclasses can safely modify

   *
   * <p>That iterator supports {@code remove()} if {@code iterable.iterator()} does. After {@code
   * remove()} is called, subsequent cycles omit the removed element, which is no longer in {@code
   * iterable}. The iterator's {@code hasNext()} method returns {@code true} until {@code iterable}
   * is empty.
   *

### API Machinery

* The Namespace API object no longer supports the deletecollection operation. ([#46407](https://github.com/kubernetes/kubernetes/pull/46407), [@liggitt](https://github.com/liggitt))

* When using client or server certificate rotation, the Kubelet will no longer wait until the initial rotation succeeds or fails before starting static pods.  This makes running self-hosted masters with rotation more predictable. ([#58930](https://github.com/kubernetes/kubernetes/pull/58930), [@smarterclayton](https://github.com/smarterclayton))

            ex0 = ex; // Defer to below where we're locked
            return;
        } finally {
            synchronized (run_thread) {
                if (run_thread != thread) {
                    /* Thread no longer the one setup for this transport --
                     * doConnect returned too late, just ignore.
                     */
                    if ((ex0 != null) && (LogStream.level >= 2)) {

      merely dropped) as soon as there are no longer any *usable*
      endpoints, rather than waiting until all of the terminating
      endpoints have terminated even when those terminating endpoints
      were not being used.
  
    - Chains for endpoints that won't be used are no longer output to
      iptables, saving a bit of memory/time/cpu. (#106373, @aojea) [SIG Network]

  - A `create` or `update` API request with duplicate items in a keyed list no longer wipes out managedFields. Examples include env var entries with the same name, or port entries with the same containerPort in a pod spec.
  - A server-side apply request that makes unrelated changes to an object which has duplicate items in a keyed list no longer fails, and leaves the existing duplicate items as-is.

        try {
            cache.getInputStream();
            fail("Expected IOException for deleted file");
        } catch (IOException e) {
            // Expected - file no longer exists
        }
    }

    @Test
    public void testFileBasedCacheCloseNonExistentFile() {
        tempFile = new File(Curl.tmpDir, "non-existent-file.tmp");
        assertFalse(tempFile.exists());

   * until a nonzero result is found; imposes "dictionary order." If the end of one iterable is
   * reached, but not the other, the shorter iterable is considered to be less than the longer one.
   * For example, a lexicographical natural ordering over integers considers {@code [] < [1] < [1,
   * 1] < [1, 2] < [2]}.
   *