// without waiting for the user's executor to run our submitted Runnable. However, this can
      // interact poorly with the reentrancy-avoiding behavior of this executor - when the operation
      // before the cancelled future completes, it will synchronously complete both the newFuture

 * {@link Multimap} itself). Prefer subtypes such as {@link ImmutableSetMultimap} or {@link
 * ImmutableListMultimap}, which have well-defined {@link #equals} semantics, thus avoiding a common
 * source of bugs and confusion.
 *
 * <p><b>Note:</b> every {@link ImmutableMultimap} offers an {@link #inverse} view, so there is no
 * need for a distinct {@code ImmutableBiMultimap} type.
 *

   *     underlying this step or, if {@code get()} throws a different kind of exception, that
   *     exception itself. To avoid hiding bugs and other unrecoverable errors, callers should
   *     prefer more specific types, avoiding {@code Throwable.class} in particular.
   * @param fallback the function to be called if this step fails with the expected exception type.

        // held
        segment.expand();
      }
      assertEquals(i, segment.table.length());
      assertEquals(liveCount, countLiveEntries(map));
      // expansion cleanup is sloppy, with a goal of avoiding unnecessary copies
      assertTrue(segment.count >= liveCount);
      assertTrue(segment.count <= originalCount);
      assertEquals(originalMap, ImmutableMap.copyOf(map));
    }
  }

        // held
        segment.expand();
      }
      assertEquals(i, segment.table.length());
      assertEquals(liveCount, countLiveEntries(map));
      // expansion cleanup is sloppy, with a goal of avoiding unnecessary copies
      assertTrue(segment.count >= liveCount);
      assertTrue(segment.count <= originalCount);
      assertEquals(originalMap, ImmutableMap.copyOf(map));
    }
  }

Just avoiding the copy of files doesn't necessarily improve things too much, but because it used the cache for that step, it can **use the cache for the next step**. For example, it could use the cache for the instruction that installs dependencies with:

   *     underlying this step or, if {@code get()} throws a different kind of exception, that
   *     exception itself. To avoid hiding bugs and other unrecoverable errors, callers should
   *     prefer more specific types, avoiding {@code Throwable.class} in particular.
   * @param fallback the function to be called if this step fails with the expected exception type.

      if (i > 1) {
        segment.expand();
      }
      assertEquals(i, segment.table.length());
      assertEquals(liveCount, countLiveEntries(map, 0));
      // expansion cleanup is sloppy, with a goal of avoiding unnecessary copies
      assertTrue(segment.count >= liveCount);
      assertTrue(segment.count <= originalCount);
      assertEquals(originalMap, ImmutableMap.copyOf(map));
    }
  }

		// of view if we simply do a GetObjectInfo() on this prefix by treating it as an object
		// We save a lot of calls over the network.
		//
		// This happens repeatedly for all objects that are created concurrently() avoiding this
		// as a List() call is an important performance improvement.
		//
		// Spark based s3a committers are a  big enough use-case to have this optimization.
		//

- Kube-proxy: on dual-stack mode, if it is not able to get the IP Family of an endpoint, logs it with level InfoV(4) instead of Warning, avoiding flooding the logs for endpoints without addresses ([#88934](https://github.com/kubernetes/kubernetes/pull/88934), [@aojea](https://github.com/aojea)) [SIG Network]