void (*deallocator)(void* data, size_t len, void* arg),
    void* deallocator_arg, TF_Status* status) {
  tensorflow::Device* device = nullptr;
  tensorflow::EagerContext* context =
      tensorflow::ContextFromInterface(tensorflow::unwrap(ctx));
  status->status = context->FindDeviceFromName(device_name, &device);
  if (!status->status.ok()) {
    deallocator(data, len, deallocator_arg);

TF_CAPI_EXPORT extern TFE_TensorHandle* TFE_NewTensorHandleFromDeviceMemory(
    TFE_Context* ctx, const char* device_name, TF_DataType, const int64_t* dims,
    int num_dims, void* data, size_t len,
    void (*deallocator)(void* data, size_t len, void* arg),
    void* deallocator_arg, TF_Status* status);

// Retrieves the address space (i.e. job, replia, task) of the local host and
// saves it in the buffer.

  TF_TString_Init(&tstr[0]);
  TF_TString_Copy(&tstr[0], test_string, sizeof(test_string) - 1);

  auto deallocator = [](void* data, size_t len, void* arg) {};
  unique_tensor_ptr t_in(TF_NewTensor(TF_STRING, nullptr, 0, &tstr[0],
                                      sizeof(tstr), deallocator, nullptr),
                         TF_DeleteTensor);

  // Create a TF_Operation with the attribute t_in

  auto dtype = TFE_TensorHandleDataType(t->tensor);
  TFE_CustomDeviceTensorHandleMethods handle_methods;
  handle_methods.num_dims = &LoggedTensorNumDims;
  handle_methods.dim = &LoggedTensorDim;
  handle_methods.deallocator = &LoggedTensorDeallocator;
  return TFE_NewCustomDeviceTensorHandle(context, logging_device_name.c_str(),
                                         dtype, t.release(), handle_methods,

// * Maybe add TF_TensorShape that encapsulates dimension info.
//
// Design decisions made:
// * Backing store for tensor memory has an associated deallocation
//   function.  This deallocation function will point to client code
//   for tensors populated by the client.  So the client can do things
//   like shadowing a numpy array.
// * We do not provide TF_OK since it is not strictly necessary and we

            lock.lock();
            return getAvailablePort();
        } finally {
            lock.unlock();
        }
    }

    /**
     * Deallocate the given port
     *
     * @param port The port to deallocate
     */
    public void deallocate(int port) {
        try {
            lock.lock();
            allocated.remove(Integer.valueOf(port));
        } finally {
            lock.unlock();
        }

    @Override
    public void apply(Project project) {
        AvailablePortAllocator allocator = project.getRootProject()
            .getPlugins()
            .apply(InternalAvailableTcpPortProviderRootPlugin.class).allocator;
        ReservedPortRange portRange = allocator.reservePortRange();
        project.getExtensions().add("portRange", portRange);
    }

import org.elasticsearch.cluster.routing.allocation.FailedShard;
import org.elasticsearch.cluster.routing.allocation.RoutingAllocation;
import org.elasticsearch.cluster.routing.allocation.allocator.BalancedShardsAllocator;
import org.elasticsearch.cluster.routing.allocation.decider.AllocationDecider;
import org.elasticsearch.cluster.routing.allocation.decider.AllocationDeciders;

    // Using @Weak is necessary to avoid retain-cycles between Node instances on iOS,
    // which would cause memory leaks when non-empty HashBiMap with cyclic Node
    // instances is deallocated.
    @Nullable Node<K, V> nextInKToVBucket;
    @Weak @Nullable Node<K, V> nextInVToKBucket;

    @Weak @Nullable Node<K, V> nextInKeyInsertionOrder;
    @Weak @Nullable Node<K, V> prevInKeyInsertionOrder;

- DRA BindingConditions: when the DRABindingConditions feature was enabled, reusing the same claim among different pods may have caused a panic in the scheduler when deallocation happens in parallel (a rare race condition). ([#137371](https://github.com/kubernetes/kubernetes/pull/137371), [@pohly](https://github.com/pohly)) [SIG Node, Scheduling and Testing]