Home
last modified time | relevance | path

Searched refs:server_node_ (Results 1 – 12 of 12) sorted by relevance

/third_party/mindspore/mindspore/ccsrc/fl/server/
Dcollective_ops_impl.cc24 server_node_ = server_node; in Initialize()
25 local_rank_ = server_node_->rank_id(); in Initialize()
72 …auto send_req_id = server_node_->CollectiveSendAsync(ps::core::NodeRole::SERVER, send_to_rank, sen… in RingAllReduce()
82 …auto recv_req_id = server_node_->CollectiveReceiveAsync(ps::core::NodeRole::SERVER, recv_from_rank… in RingAllReduce()
83 if (!server_node_->CollectiveWait(recv_req_id, kCollectiveCommTimeout)) { in RingAllReduce()
98 if (!server_node_->Wait(send_req_id, kCollectiveCommTimeout)) { in RingAllReduce()
110 …auto send_req_id = server_node_->CollectiveSendAsync(ps::core::NodeRole::SERVER, send_to_rank, sen… in RingAllReduce()
119 …auto recv_req_id = server_node_->CollectiveReceiveAsync(ps::core::NodeRole::SERVER, recv_from_rank… in RingAllReduce()
120 if (!server_node_->CollectiveWait(recv_req_id, kCollectiveCommTimeout)) { in RingAllReduce()
129 if (!server_node_->Wait(send_req_id, kCollectiveCommTimeout)) { in RingAllReduce()
[all …]
Dserver.cc150 server_node_ = std::make_shared<ps::core::ServerNode>(); in InitCluster()
151 MS_EXCEPTION_IF_NULL(server_node_); in InitCluster()
167 MS_EXCEPTION_IF_NULL(server_node_); in InitCommunicatorWithServer()
168 …communicator_with_server_ = server_node_->GetOrCreateTcpComm(scheduler_ip_, static_cast<int16_t>(s… in InitCommunicatorWithServer()
176 MS_EXCEPTION_IF_NULL(server_node_); in InitCommunicatorWithWorker()
189 …auto http_comm = server_node_->GetOrCreateHttpComm(server_node_->BoundIp(), http_port_, task_execu… in InitCommunicatorWithWorker()
293 MS_EXCEPTION_IF_NULL(server_node_); in RegisterCommCallbacks()
303 iteration_->RegisterEventCallback(server_node_); in RegisterCommCallbacks()
310 if (!server_node_->InitFollowerScaler()) { in RegisterCommCallbacks()
315 server_node_->RegisterFollowerScalerBarrierBeforeScaleOut("ServerPipeline", in RegisterCommCallbacks()
[all …]
Diteration.cc56 server_node_ = server_node; in RegisterEventCallback()
122 MS_ERROR_IF_NULL_WO_RET_VAL(server_node_); in MoveToNextIteration()
123 if (server_node_->rank_id() == kLeaderServerRank) { in MoveToNextIteration()
139 … MS_LOG(ERROR) << "Server " << server_node_->rank_id() << " notifying the leader server failed."; in MoveToNextIteration()
147 MS_ERROR_IF_NULL_WO_RET_VAL(server_node_); in SetIterationRunning()
148 if (server_node_->rank_id() == kLeaderServerRank) { in SetIterationRunning()
150 server_node_->BroadcastEvent(static_cast<uint32_t>(ps::CustomEvent::kIterationRunning)); in SetIterationRunning()
160 MS_ERROR_IF_NULL_WO_RET_VAL(server_node_); in SetIterationCompleted()
161 if (server_node_->rank_id() == kLeaderServerRank) { in SetIterationCompleted()
163 server_node_->BroadcastEvent(static_cast<uint32_t>(ps::CustomEvent::kIterationCompleted)); in SetIterationCompleted()
[all …]
Ddistributed_metadata_store.cc27 server_node_ = server_node; in Initialize()
28 local_rank_ = server_node_->rank_id(); in Initialize()
154 if (server_node_ == nullptr) { in ReInitForScaling()
159 local_rank_ = server_node_->rank_id(); in ReInitForScaling()
160 server_num_ = IntToUint(server_node_->server_num()); in ReInitForScaling()
Dcollective_ops_impl.h54 CollectiveOpsImpl() : server_node_(nullptr), local_rank_(0), server_num_(0) {} in CollectiveOpsImpl()
67 std::shared_ptr<ps::core::ServerNode> server_node_; variable
Ddistributed_metadata_store.h69 : server_node_(nullptr), in DistributedMetadataStore()
92 std::shared_ptr<ps::core::ServerNode> server_node_; variable
Ddistributed_count_service.cc28 server_node_ = server_node; in Initialize()
29 local_rank_ = server_node_->rank_id(); in Initialize()
174 if (server_node_ == nullptr) { in ReInitForScaling()
179 local_rank_ = server_node_->rank_id(); in ReInitForScaling()
180 server_num_ = IntToUint(server_node_->server_num()); in ReInitForScaling()
Dserver.h66 : server_node_(nullptr), in Server()
152 std::shared_ptr<ps::core::ServerNode> server_node_; variable
Diteration.h127 server_node_(nullptr), in Iteration()
198 std::shared_ptr<ps::core::ServerNode> server_node_; variable
Ddistributed_count_service.h110 std::shared_ptr<ps::core::ServerNode> server_node_; variable
/third_party/mindspore/mindspore/ccsrc/ps/
Dparameter_server.cc29 server_node_ = std::make_shared<core::ServerNode>(); in Run()
37 server_node_->Start(); in Run()
38 PSContext::instance()->SetPSRankId(server_node_->rank_id()); in Run()
42 server_node_->Finish(); in Run()
43 if (!server_node_->Stop()) { in Run()
57 server_node_->set_handler(*handler_); in Init()
58 server_node_->RegisterEventCallback(core::ClusterEvent::SCHEDULER_TIMEOUT, [this]() { in Init()
62 server_node_->RegisterEventCallback(core::ClusterEvent::NODE_TIMEOUT, [this]() { in Init()
125 …std::make_shared<kernel::ps::SparseApplyAdamPSKernel>(server_node_->rank_id(), pserver_num_, worke… in InitOptimInputsShape()
130 …std::make_shared<kernel::ps::SparseApplyLazyAdamPSKernel>(server_node_->rank_id(), pserver_num_, w… in InitOptimInputsShape()
[all …]
Dparameter_server.h87 server_node_(nullptr) {} in ParameterServer()
177 std::shared_ptr<core::ServerNode> server_node_; variable