func newBatchJobPool(ctx context.Context, o ObjectLayer, workers int) *BatchJobPool {
	jpool := &BatchJobPool{
		ctx:          ctx,
		objLayer:     o,
		jobCh:        make(chan *BatchJobRequest, 10000),
		workerKillCh: make(chan struct{}, workers),
		jobCancelers: make(map[string]context.CancelFunc),
	}
	jpool.ResizeWorkers(workers)

	randomWait := func() time.Duration {

	localDrives := cloneDrives(globalLocalDrivesMap)
	globalLocalDrivesMu.RUnlock()

	errs := make([]error, len(localDrives))
	infos := make([]DiskInfo, len(localDrives))
	for i, drive := range localDrives {
		infos[i], errs[i] = drive.DiskInfo(GlobalContext, DiskInfoOptions{})
	}
	infoMaps := make(map[string]madmin.DiskMetrics)
	for i := range infos {

				}
			}
			if err := w.Close(); err != nil {
				t.Fatal(err)
			}
		}
	}
	// 16k-1 records shouldn't make a zip64:
	t.Run("uint16max-1_NoZip64", func(t *testing.T) {
		t.Parallel()
		if generatesZip64(t, gen(0xfffe)) {
			t.Error("unexpected zip64")
		}
	})
	// 16k records should make a zip64:
	t.Run("uint16max_Zip64", func(t *testing.T) {
		t.Parallel()
		if !generatesZip64(t, gen(0xffff)) {

	b.ReportAllocs()
	b.ResetTimer()
	for i := 0; i < b.N; i++ {
		v.MarshalMsg(nil)
	}
}

func BenchmarkAppendMsgBucketReplicationStat(b *testing.B) {
	v := BucketReplicationStat{}
	bts := make([]byte, 0, v.Msgsize())
	bts, _ = v.MarshalMsg(bts[0:0])
	b.SetBytes(int64(len(bts)))
	b.ReportAllocs()
	b.ResetTimer()
	for i := 0; i < b.N; i++ {
		bts, _ = v.MarshalMsg(bts[0:0])
	}
}

	f, err := OsOpenFile(name, readMode, 0o666)
	if err != nil {
		return nil, nil, err
	}
	defer f.Close()

	st, err := f.Stat()
	if err != nil {
		return nil, nil, err
	}

	dst := make([]byte, st.Size())
	_, err = io.ReadFull(f, dst)
	return dst, st, err
}

// ReadFile reads the named file and returns the contents.
// A successful call returns err == nil, not err == EOF.

/tensorflow/tools/tf_sig_build_dockerfiles/. However, since
hermetic CUDA and hermetic Python is now available for Tensorflow, a lot of the
requirements installed on the original container can be removed to reduce the
footprint of the container and make it more reusable across different ML

    assertTrue(iter.hasNext());
    assertTrue(iter.hasNext());
    assertTrue(iter.hasNext());
    assertEquals(1, (int) iter.next());

    assertFalse(iter.hasNext());

    // Make sure computeNext() doesn't get invoked again
    assertFalse(iter.hasNext());

    assertThrows(NoSuchElementException.class, iter::next);
  }

  public void testDefaultBehaviorOfPeek() {
    /*

    assertTrue(iter.hasNext());
    assertTrue(iter.hasNext());
    assertTrue(iter.hasNext());
    assertEquals(1, (int) iter.next());

    assertFalse(iter.hasNext());

    // Make sure computeNext() doesn't get invoked again
    assertFalse(iter.hasNext());

    assertThrows(NoSuchElementException.class, iter::next);
  }

  public void testDefaultBehaviorOfPeek() {
    /*

  private final ElementOrder<N> nodeOrder;
  private final ElementOrder<E> edgeOrder;

  final MapIteratorCache<N, NetworkConnections<N, E>> nodeConnections;

  // We could make this a Map<E, EndpointPair<N>>. It would make incidentNodes(edge) slightly
  // faster, but also make Networks consume 5 to 20+% (increasing with average degree) more memory.
  final MapIteratorCache<E, N> edgeToReferenceNode; // referenceNode == source if directed

  private final ElementOrder<N> nodeOrder;
  private final ElementOrder<E> edgeOrder;

  final MapIteratorCache<N, NetworkConnections<N, E>> nodeConnections;

  // We could make this a Map<E, EndpointPair<N>>. It would make incidentNodes(edge) slightly
  // faster, but also make Networks consume 5 to 20+% (increasing with average degree) more memory.
  final MapIteratorCache<E, N> edgeToReferenceNode; // referenceNode == source if directed