From a0df28357f5b9a6a17bb3a08ee1ca601e078165b Mon Sep 17 00:00:00 2001 From: chendongsheng Date: Thu, 8 Jul 2021 11:16:49 +0800 Subject: [PATCH] fixed state --- mindspore/ccsrc/ps/core/abstract_node.cc | 1 - mindspore/ccsrc/ps/core/scheduler_node.cc | 2 +- 2 files changed, 1 insertion(+), 2 deletions(-) diff --git a/mindspore/ccsrc/ps/core/abstract_node.cc b/mindspore/ccsrc/ps/core/abstract_node.cc index 38e923a9e1e..60b8ffdd5be 100644 --- a/mindspore/ccsrc/ps/core/abstract_node.cc +++ b/mindspore/ccsrc/ps/core/abstract_node.cc @@ -586,7 +586,6 @@ void AbstractNode::ProcessSendMetadata(std::shared_ptr conn, std: std::lock_guard lock(client_mutex_); connected_nodes_.clear(); - MS_LOG(INFO) << "The current cluster state:" << CommUtil::ClusterStateToString(current_cluster_state_); } void AbstractNode::ProcessFinish(std::shared_ptr conn, std::shared_ptr meta, diff --git a/mindspore/ccsrc/ps/core/scheduler_node.cc b/mindspore/ccsrc/ps/core/scheduler_node.cc index 3f2911e9b40..337c197d2b8 100644 --- a/mindspore/ccsrc/ps/core/scheduler_node.cc +++ b/mindspore/ccsrc/ps/core/scheduler_node.cc @@ -148,7 +148,6 @@ void SchedulerNode::ProcessRegister(std::shared_ptr server, std::shar is_ready_ = true; MS_LOG(INFO) << "There are " << node_manager_.worker_num() << " workers and " << node_manager_.server_num() << " servers registered to scheduer, so the scheduler send meta data to worker/server."; - node_manager_.UpdateClusterState(ClusterState::CLUSTER_READY); if (node_manager_.GetClusterState() == ClusterState::CLUSTER_SCALE_IN) { auto nodes = node_manager_.nodes_info(); for (const auto &id : scale_in_node_ids_) { @@ -163,6 +162,7 @@ void SchedulerNode::ProcessRegister(std::shared_ptr server, std::shar auto client = GetOrCreateClient(kvs.second); SendMetadata(client, kvs.second.rank_id_); } + node_manager_.UpdateClusterState(ClusterState::CLUSTER_READY); wait_start_cond_.notify_all(); } }