if dswp.podPreviouslyProcessed(uniquePodName) {
		return
	}

	allVolumesAdded := true
	mounts, devices, seLinuxContainerContexts := util.GetPodVolumeNames(pod)

	// Process volume spec for each volume defined in pod
	for _, podVolume := range pod.Spec.Volumes {
		if !mounts.Has(podVolume.Name) && !devices.Has(podVolume.Name) {
			// Volume is not used in the pod, ignore it.

		t.Run(test.name, func(t *testing.T) {
			mounts, devices, contexts := GetPodVolumeNames(test.pod)
			if !mounts.Equal(test.expectedMounts) {
				t.Errorf("Expected mounts: %q, got %q", sets.List[string](mounts), sets.List[string](test.expectedMounts))
			}
			if !devices.Equal(test.expectedDevices) {
				t.Errorf("Expected devices: %q, got %q", sets.List[string](devices), sets.List[string](test.expectedDevices))
			}

  %5:2 = tf_device.replicate([%2#0, %3#0] as %r0: tensor<3x3x1x32xf32>, [%2#1, %3#1] as %r1: tensor<3x3x1x32xf32>)
      {n = 2 : i32, devices = {TPU_REPLICATED_CORE_0 = ["/device:TPU:0", "/device:TPU:1"]}} {
    // CHECK: "tf.TPUExecute"(%[[R0]], %[[R1]], %[[COMPILE]]#1)
    %execute = "tf_device.launch"() ({

					t.Fatalf("expected error message `%s` but got `%v`", tc.expectedErrMsg, err)
				}
				return
			}
			// otherwise validate the devices
			if err != nil {
				t.Fatal(err)
			}
			assert.Equal(t, tc.expectedDevices, blkVolumes, "devices of container %+v", tc.container)
		})
	}

// product of possible devices and inference datatypes. For example, we might
// raise a chunk of sequential operations from a block all having attributes
// `{ tac.device = "GPU", tac.inference_type = "FLOAT"}` to a function
// with the matching attributes. Assumed is that device type "CPU"
// is the only device that is allowed to call other devices. I.e. ancestors of a

  float* m_float = static_cast<float*>(TF_TensorData(m_data));
  TF_DeviceList* devices = TFE_ContextListDevices(ctx, status);
  CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  int num_devices = TF_DeviceListCount(devices);
  for (int d = 0; d < num_devices; ++d) {
    const char* name = TF_DeviceListName(devices, d, status);
    CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

// associative and commutative, and produce output with the same shape as the
// input. The optimizer may replace an aggregate op taking input from multiple
// devices with a tree of aggregate ops that aggregate locally within each
// device (and possibly within groups of nearby devices) before communicating.
TF_CAPI_EXPORT extern void TF_OpDefinitionBuilderSetIsAggregate(
    TF_OpDefinitionBuilder* builder, bool is_aggregate);

// Destroy an options object.
TF_CAPI_EXPORT extern void TFE_DeleteContextOptions(TFE_ContextOptions*);

// "Context" under which operations/functions are executed. It encapsulates
// things like the available devices, resource manager etc.
// TFE_Context must outlive all tensor handles created using it. In other
// words, TFE_DeleteContext() must be called after all tensor handles have
// been deleted (with TFE_DeleteTensorHandle).
//

// getExpectedVolumes returns a list of volumes that must be mounted in order to
// consider the volume setup step for this pod satisfied.
func getExpectedVolumes(pod *v1.Pod) []string {
	mounts, devices, _ := util.GetPodVolumeNames(pod)
	return mounts.Union(devices).UnsortedList()
}

// getExtraSupplementalGid returns the value of an extra supplemental GID as
// defined by an annotation on a volume and a boolean indicating whether the

module attributes {tf.devices = {"/job:tpu_host_worker/replica:0/task:0/device:CPU:0", "/job:tpu_host_worker/replica:0/task:0/device:TPU:0", "/job:tpu_host_worker/replica:0/task:0/device:TPU:1", "/job:tpu_host_worker/replica:0/task:0/device:TPU_SYSTEM:0", "/job:tpu_host_worker/replica:0/task:1/device:CPU:0", "/job:tpu_host_worker/replica:0/task:1/device:TPU:0", "/job:tpu_host_worker/replica:0/task:1/device:TPU:1", "/job:tpu_host_worker/replica:0/task:1/device:TPU_SYSTEM:0", "/job:tpu_host_wor...