1 /* Copyright 2018 The TensorFlow Authors. All Rights Reserved. 2 3 Licensed under the Apache License, Version 2.0 (the "License"); 4 you may not use this file except in compliance with the License. 5 You may obtain a copy of the License at 6 7 http://www.apache.org/licenses/LICENSE-2.0 8 9 Unless required by applicable law or agreed to in writing, software 10 distributed under the License is distributed on an "AS IS" BASIS, 11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12 See the License for the specific language governing permissions and 13 limitations under the License. 14 ==============================================================================*/ 15 #ifndef TENSORFLOW_CORE_DISTRIBUTED_RUNTIME_COLLECTIVE_RMA_DISTRIBUTED_H_ 16 #define TENSORFLOW_CORE_DISTRIBUTED_RUNTIME_COLLECTIVE_RMA_DISTRIBUTED_H_ 17 18 #include "tensorflow/core/common_runtime/collective_rma_local.h" 19 #include "tensorflow/core/framework/cancellation.h" 20 #include "tensorflow/core/framework/tensor.h" 21 #include "tensorflow/core/platform/unbounded_work_queue.h" 22 23 namespace tensorflow { 24 class WorkerCacheInterface; 25 26 // Extend CollectiveRemoteAccessLocal with access to remote peers. 27 class CollectiveRemoteAccessDistributed : public CollectiveRemoteAccessLocal { 28 public: CollectiveRemoteAccessDistributed(const DeviceMgr * dev_mgr,DeviceResolverInterface * dev_resolver,std::shared_ptr<UnboundedWorkQueue> work_queue,WorkerCacheInterface * worker_cache,int64 step_id,string task_name)29 CollectiveRemoteAccessDistributed( 30 const DeviceMgr* dev_mgr, DeviceResolverInterface* dev_resolver, 31 std::shared_ptr<UnboundedWorkQueue> work_queue, 32 WorkerCacheInterface* worker_cache, int64 step_id, string task_name) 33 : CollectiveRemoteAccessLocal(dev_mgr, dev_resolver, step_id), 34 worker_cache_(worker_cache), 35 work_queue_(std::move(work_queue)), 36 task_name_(std::move(task_name)) {} 37 ~CollectiveRemoteAccessDistributed()38 ~CollectiveRemoteAccessDistributed() override {} 39 40 void RecvFromPeer(const string& peer_device, const string& peer_task, 41 bool peer_is_local, const string& key, Device* to_device, 42 DeviceContext* to_device_ctx, 43 const AllocatorAttributes& to_alloc_attr, Tensor* to_tensor, 44 const DeviceLocality& client_locality, 45 int dev_to_dev_stream_index, 46 CancellationManager* cancellation_manager, 47 const StatusCallback& done) override; 48 49 void CheckPeerHealth(const string& peer_task, int64 timeout_in_ms, 50 const StatusCallback& done) override; 51 52 void StartAbort(const Status& s) override; 53 54 protected: 55 WorkerCacheInterface* worker_cache_; // Not owned 56 // Ownership of `work_queue_` is shared between `this` and 57 // `CollectiveExecutorMgr`. 58 std::shared_ptr<UnboundedWorkQueue> work_queue_; 59 CancellationManager abortion_cancel_mgr_; 60 string task_name_; 61 }; 62 63 } // namespace tensorflow 64 #endif // TENSORFLOW_CORE_DISTRIBUTED_RUNTIME_COLLECTIVE_RMA_DISTRIBUTED_H_ 65