(`tensorflow/tensorflow:devel` and `tensorflow/tensorflow:devel-gpu` are no
    longer supported for development). Use TF SIG Build Dockerfiles in
    development to avoid installing the packages directly on your system (in
    which case remember to change the directory from `/root` to `/tensorflow`
    once you get into the running container so `bazel` can find the `tensorflow`
    workspace).

- DaemonSet controller is now using backoff algorithm to avoid hot loops fighting with kubelet on pod recreation when a particular DaemonSet is misconfigured. ([#65309](https://github.com/kubernetes/kubernetes/pull/65309), [@tnozicka](https://github.com/tnozicka))

					ObjectOptions{
						DeletePrefix:       true, // use prefix delete to delete all versions at once.
						DeletePrefixObject: true, // use prefix delete on exact object (this is an optimization to avoid fan-out calls)
						NoAuditLog:         true,
					},
				)
				stopFn(0, err)
				auditLogDecom(ctx, "DecomDeleteObject", bi.Name, entry.name, "", err)
				if err != nil {
					decomLogIf(ctx, err)
				}

   * is {@code 60}, and the concurrency level is {@code 8}, then eight segments are created, each
   * having a hash table of size eight. Providing a large enough estimate at construction time
   * avoids the need for expensive resizing operations later, but setting this value unnecessarily
   * high wastes memory.
   *
   * @return this {@code CacheBuilder} instance (for chaining)

* Improved user experience for environments without a public internet connection and users using other CRI runtimes than Docker
* Support for structured configuration for the kubelet, which avoids the need to modify the systemd drop-in file
* Many improvements to the upgrade process and other bug fixes

### SIG Instrumentation

     * a given entry might not include null. The right fix for the header problems is probably to
     * define a separate MultimapLink interface with a separate "header" implementation, which
     * hopefully could avoid implementing Entry or ValueSetLink at all. (But note that that approach
     * requires us to define extra classes -- unfortunate under Android.) *Then* we could consider

      } else {
        result.set(i, getAt(currentStripe));
        previousStripe = currentStripe;
      }
    }
    /*
     * Note that the returned Iterable holds references to the returned stripes, to avoid
     * error-prone code like:
     *
     * Striped<Lock> stripedLock = Striped.lazyWeakXXX(...)'
     * Iterable<Lock> locks = stripedLock.bulkGet(keys);
     * for (Lock lock : locks) {
     *   lock.lock();

{!> ../../docs_src/python_types/tutorial009b.py!}
```

////

#### Using `Union` or `Optional`

If you are using a Python version below 3.10, here's a tip from my very **subjective** point of view:

* 🚨 Avoid using `Optional[SomeType]`
* Instead ✨ **use `Union[SomeType, None]`** ✨.

@SuppressWarnings("unused") // Nested enums used reflectively in setUp.
public class EnumsBenchmark {

  @Param({"Small", "Medium", "Large"})
  String enumSize;

  @Param({"0.2", "0.8"})
  float hitRate;

  // We could avoid the raw type here by initializing this with a ternary (? SmallEnum.class : ...).
  // However, we end up needing a raw type in getIfPresent, as discussed there.
  @SuppressWarnings("rawtypes")
  private Class<? extends Enum> enumType;

  optional string devicePath = 2;
}

// AvoidPods describes pods that should avoid this node. This is the value for a
// Node annotation with key scheduler.alpha.kubernetes.io/preferAvoidPods and
// will eventually become a field of NodeStatus.
message AvoidPods {
  // Bounded-sized list of signatures of pods that should avoid this node, sorted
  // in timestamp order from oldest to newest. Size of the slice is unspecified.