* returns a <b>(live) unmodifiable view</b>; see <a
 * href="https://github.com/google/guava/wiki/GraphsExplained#accessor-behavior">the external
 * documentation</a> for details.
 *
 * <p>Examples of use:
 *
 * <pre>{@code
 * // Building a mutable value graph
 * MutableValueGraph<String, Double> graph =
 *     ValueGraphBuilder.undirected().allowsSelfLoops(true).build();
 * graph.putEdgeValue("San Francisco", "San Francisco", 0.0);

   *
   * <pre>{@code
   * static final ImmutableClassToInstanceMap<Handler> HANDLERS =
   *     new ImmutableClassToInstanceMap.Builder<Handler>()
   *         .put(FooHandler.class, new FooHandler())
   *         .put(BarHandler.class, new SubBarHandler())
   *         .put(Handler.class, new QuuxHandler())
   *         .build();
   * }</pre>
   *

   *
   * <p>Suppose {@code x} is a queue known to contain only strings. The following code can be used
   * to dump the queue into a newly allocated array of {@code String}:
   *
   * <pre>
   *     String[] y = x.toArray(new String[0]);</pre>
   *
   * <p>Note that {@code toArray(new Object[0])} is identical in function to {@code toArray()}.
   *

   *
   * <p><b>Note:</b> on Java 8+, it is usually better to use streams. For example:
   *
   * <pre>{@code
   * import static com.google.common.collect.ImmutableMap.toImmutableMap;
   * ...
   * ImmutableMap<Color, String> colorNames =
   *     allColors.stream().collect(toImmutableMap(c -> c, c -> c.toString()));
   * }</pre>
   *
   * <p>Streams provide a more standard and flexible API and the lambdas make it clear what the keys

   *
   * <p><b>Note:</b> on Java 8+, it is usually better to use streams. For example:
   *
   * <pre>{@code
   * import static com.google.common.collect.ImmutableMap.toImmutableMap;
   * ...
   * ImmutableMap<Color, String> colorNames =
   *     allColors.stream().collect(toImmutableMap(c -> c, c -> c.toString()));
   * }</pre>
   *
   * <p>Streams provide a more standard and flexible API and the lambdas make it clear what the keys

	if !s.IsValid(w, r) {
		return
	}
	volume := r.Form.Get(storageRESTVolume)

	var preq ReadPartsReq
	if err := msgp.Decode(r.Body, &preq); err != nil {
		s.writeErrorResponse(w, err)
		return
	}

	done := keepHTTPResponseAlive(w)
	infos, err := s.getStorage().ReadParts(r.Context(), volume, preq.Paths...)
	done(nil)
	if err != nil {
		s.writeErrorResponse(w, err)
		return
	}

   * {@code ClosingFuture}.
   *
   * <p>Example usage:
   *
   * <pre>{@code
   * ClosingFuture<List<Row>> rowsFuture =
   *     queryFuture.transform((closer, result) -> result.getRows(), executor);
   * }</pre>
   *
   * <p>When selecting an executor, note that {@code directExecutor} is dangerous in some cases. See

TFCI_ARTIFACT_FINAL_PYPI_ENABLE=1
TFCI_ARTIFACT_LATEST_GCS_URI="gs://tensorflow/versions/latest/"
TFCI_ARTIFACT_STAGING_GCS_ENABLE=1
TFCI_ARTIFACT_STAGING_GCS_URI="gs://tensorflow-ci-staging/staging/versions/${_LOUHI_EXECUTION_ID:-$(git rev-parse HEAD)}/"

#### Endpoints

     * The pool will routinely try to pre-emptively open connections to satisfy this minimum.
     * Connections will still be closed if they idle beyond the keep-alive but will be replaced.
     */
    @JvmField val minimumConcurrentCalls: Int = 0,
    /** How long to wait to retry pre-emptive connection attempts that fail. */
    @JvmField val backoffDelayMillis: Long = 60 * 1000,