diff --git a/mindspore/ccsrc/ps/CMakeLists.txt b/mindspore/ccsrc/ps/CMakeLists.txt index 091b77193e..ab34d8ac73 100644 --- a/mindspore/ccsrc/ps/CMakeLists.txt +++ b/mindspore/ccsrc/ps/CMakeLists.txt @@ -12,7 +12,6 @@ if(NOT (ENABLE_CPU AND (ENABLE_D OR ENABLE_GPU))) list(REMOVE_ITEM _PS_SRC_FILES "core/tcp_client.cc") list(REMOVE_ITEM _PS_SRC_FILES "core/tcp_message_handler.cc") list(REMOVE_ITEM _PS_SRC_FILES "core/tcp_server.cc") - list(REMOVE_ITEM _PS_SRC_FILES "core/cluster_metadata.cc") list(REMOVE_ITEM _PS_SRC_FILES "core/node.cc") list(REMOVE_ITEM _PS_SRC_FILES "core/node_manager.cc") list(REMOVE_ITEM _PS_SRC_FILES "ps_cache/ps_cache_manager.cc") diff --git a/mindspore/ccsrc/ps/core/cluster_metadata.cc b/mindspore/ccsrc/ps/core/cluster_metadata.cc index 7f704fab63..6d8bea0dc0 100644 --- a/mindspore/ccsrc/ps/core/cluster_metadata.cc +++ b/mindspore/ccsrc/ps/core/cluster_metadata.cc @@ -32,9 +32,6 @@ void ClusterMetadata::Init(const uint32_t &worker_num, const uint32_t &server_nu const uint16_t &scheduler_port) { worker_num_ = worker_num; server_num_ = server_num; - if (!CommUtil::CheckIp(scheduler_host)) { - MS_LOG(EXCEPTION) << "The scheduler_host:" << scheduler_host << " is illegal!"; - } scheduler_host_ = std::make_unique(scheduler_host); scheduler_port_ = scheduler_port; } diff --git a/mindspore/ccsrc/ps/core/cluster_metadata.h b/mindspore/ccsrc/ps/core/cluster_metadata.h index 2039aba0b7..0fbf039a9d 100644 --- a/mindspore/ccsrc/ps/core/cluster_metadata.h +++ b/mindspore/ccsrc/ps/core/cluster_metadata.h @@ -23,7 +23,6 @@ #include #include "utils/log_adapter.h" -#include "ps/core/comm_util.h" namespace mindspore { namespace ps { diff --git a/mindspore/ccsrc/ps/core/http_client.h b/mindspore/ccsrc/ps/core/http_client.h index efef6d8ac2..3da0ea378f 100644 --- a/mindspore/ccsrc/ps/core/http_client.h +++ b/mindspore/ccsrc/ps/core/http_client.h @@ -45,7 +45,6 @@ namespace mindspore { namespace ps { namespace core { - enum class HttpMethod { HM_GET = 1 << 0, HM_POST = 1 << 1 }; enum class Status : int { diff --git a/mindspore/ccsrc/ps/core/scheduler_node.cc b/mindspore/ccsrc/ps/core/scheduler_node.cc index 40dcc9eda0..200f468b8a 100644 --- a/mindspore/ccsrc/ps/core/scheduler_node.cc +++ b/mindspore/ccsrc/ps/core/scheduler_node.cc @@ -19,7 +19,6 @@ namespace mindspore { namespace ps { namespace core { - SchedulerNode::~SchedulerNode() { MS_LOG(INFO) << "Stop scheduler node!"; Stop(); diff --git a/mindspore/ccsrc/ps/core/scheduler_node.h b/mindspore/ccsrc/ps/core/scheduler_node.h index c3d6e4dfb8..8a682770da 100644 --- a/mindspore/ccsrc/ps/core/scheduler_node.h +++ b/mindspore/ccsrc/ps/core/scheduler_node.h @@ -36,7 +36,6 @@ namespace mindspore { namespace ps { namespace core { - class SchedulerNode : public Node { public: SchedulerNode() : server_(nullptr), scheduler_thread_(nullptr), update_state_thread_(nullptr) {} diff --git a/mindspore/ccsrc/ps/core/tcp_client.cc b/mindspore/ccsrc/ps/core/tcp_client.cc index 25b0c9cbb9..7a8094be4c 100644 --- a/mindspore/ccsrc/ps/core/tcp_client.cc +++ b/mindspore/ccsrc/ps/core/tcp_client.cc @@ -30,8 +30,6 @@ #include #include -#include "ps/core/comm_util.h" - namespace mindspore { namespace ps { namespace core { @@ -335,7 +333,6 @@ void TcpClient::StartTimer(const uint32_t &time) { void TcpClient::set_timer_callback(const OnTimer &timer) { on_timer_callback_ = timer; } const event_base &TcpClient::eventbase() { return *event_base_; } - } // namespace core } // namespace ps } // namespace mindspore diff --git a/mindspore/ccsrc/ps/core/tcp_client.h b/mindspore/ccsrc/ps/core/tcp_client.h index ba22e15a86..57761b266a 100644 --- a/mindspore/ccsrc/ps/core/tcp_client.h +++ b/mindspore/ccsrc/ps/core/tcp_client.h @@ -34,6 +34,7 @@ #include "ps/core/cluster_metadata.h" #include "utils/convert_utils_base.h" +#include "ps/core/comm_util.h" namespace mindspore { namespace ps { @@ -99,7 +100,6 @@ class TcpClient { std::atomic is_stop_; std::atomic is_connected_; }; - } // namespace core } // namespace ps } // namespace mindspore diff --git a/mindspore/ccsrc/ps/core/tcp_server.cc b/mindspore/ccsrc/ps/core/tcp_server.cc index aab6ee1b3b..ed497a4d4a 100644 --- a/mindspore/ccsrc/ps/core/tcp_server.cc +++ b/mindspore/ccsrc/ps/core/tcp_server.cc @@ -29,8 +29,6 @@ #include #include -#include "ps/core/comm_util.h" - namespace mindspore { namespace ps { namespace core { @@ -445,7 +443,6 @@ int TcpServer::ConnectionNum() const { return connections_.size(); } const std::map> &TcpServer::Connections() const { return connections_; } void TcpServer::SetMessageCallback(const OnServerReceiveMessage &cb) { message_callback_ = cb; } - } // namespace core } // namespace ps } // namespace mindspore diff --git a/mindspore/ccsrc/ps/core/tcp_server.h b/mindspore/ccsrc/ps/core/tcp_server.h index 3c38a6dd40..d0a3be7cb0 100644 --- a/mindspore/ccsrc/ps/core/tcp_server.h +++ b/mindspore/ccsrc/ps/core/tcp_server.h @@ -37,6 +37,7 @@ #include "ps/core/tcp_message_handler.h" #include "ps/core/cluster_metadata.h" #include "utils/convert_utils_base.h" +#include "ps/core/comm_util.h" namespace mindspore { namespace ps { diff --git a/mindspore/ccsrc/ps/core/worker_node.cc b/mindspore/ccsrc/ps/core/worker_node.cc index 774220aaa5..11216a7ce8 100644 --- a/mindspore/ccsrc/ps/core/worker_node.cc +++ b/mindspore/ccsrc/ps/core/worker_node.cc @@ -15,6 +15,7 @@ */ #include "ps/core/worker_node.h" +#include "ps/core/comm_util.h" namespace mindspore { namespace ps { diff --git a/mindspore/ccsrc/ps/parameter_server.cc b/mindspore/ccsrc/ps/parameter_server.cc index a2ca6646e9..931e9d66ce 100644 --- a/mindspore/ccsrc/ps/parameter_server.cc +++ b/mindspore/ccsrc/ps/parameter_server.cc @@ -23,9 +23,6 @@ void ParameterServer::Run(const FuncGraphPtr &func_graph) { MS_LOG(INFO) << "PServer starts connecting to scheduler and workers..."; server_node_ = std::make_shared(); - core::ClusterMetadata::instance()->Init( - PSContext::instance()->initial_worker_num(), PSContext::instance()->initial_server_num(), - PSContext::instance()->scheduler_host(), PSContext::instance()->scheduler_port()); MS_LOG(INFO) << "PServer connected successfully."; if (!PSContext::instance()->is_server()) { MS_LOG(INFO) << "This is not the Server node."; diff --git a/mindspore/ccsrc/ps/parameter_server.h b/mindspore/ccsrc/ps/parameter_server.h index a8fd5f2277..dd56d6ec10 100644 --- a/mindspore/ccsrc/ps/parameter_server.h +++ b/mindspore/ccsrc/ps/parameter_server.h @@ -63,7 +63,6 @@ namespace mindspore { namespace ps { - class ParameterServer { public: static ParameterServer &GetInstance() { diff --git a/mindspore/ccsrc/ps/ps_context.cc b/mindspore/ccsrc/ps/ps_context.cc index 1c11423bd5..039f8f792a 100644 --- a/mindspore/ccsrc/ps/ps_context.cc +++ b/mindspore/ccsrc/ps/ps_context.cc @@ -52,6 +52,7 @@ void PSContext::SetPSEnable(bool enabled) { server_num_ = std::strtol(common::GetEnv(kEnvPServerNum).c_str(), nullptr, 10); scheduler_host_ = common::GetEnv(kEnvSchedulerHost); scheduler_port_ = std::strtol(common::GetEnv(kEnvSchedulerPort).c_str(), nullptr, 10); + core::ClusterMetadata::instance()->Init(worker_num_, server_num_, scheduler_host_, scheduler_port_); } else { MS_LOG(INFO) << "PS mode is disabled."; is_worker_ = false; diff --git a/mindspore/ccsrc/ps/ps_context.h b/mindspore/ccsrc/ps/ps_context.h index e6106696ce..1f474038a4 100644 --- a/mindspore/ccsrc/ps/ps_context.h +++ b/mindspore/ccsrc/ps/ps_context.h @@ -20,6 +20,7 @@ #include #include #include "ps/constants.h" +#include "ps/core/cluster_metadata.h" namespace mindspore { namespace ps { diff --git a/mindspore/ccsrc/ps/scheduler.cc b/mindspore/ccsrc/ps/scheduler.cc index 2af96f9741..acb1ca785c 100755 --- a/mindspore/ccsrc/ps/scheduler.cc +++ b/mindspore/ccsrc/ps/scheduler.cc @@ -19,9 +19,7 @@ namespace mindspore { namespace ps { void Scheduler::Run() { - core::ClusterMetadata::instance()->Init( - PSContext::instance()->initial_worker_num(), PSContext::instance()->initial_server_num(), - PSContext::instance()->scheduler_host(), PSContext::instance()->scheduler_port()); + MS_LOG(INFO) << "Start scheduler."; scheduler_node_.Start(); scheduler_node_.Finish(); scheduler_node_.Stop(); diff --git a/mindspore/ccsrc/ps/worker.cc b/mindspore/ccsrc/ps/worker.cc index e56cbeac77..28dacb4247 100644 --- a/mindspore/ccsrc/ps/worker.cc +++ b/mindspore/ccsrc/ps/worker.cc @@ -22,9 +22,6 @@ namespace ps { void Worker::Run() { std::lock_guard lock(running_mutex_); - core::ClusterMetadata::instance()->Init( - PSContext::instance()->initial_worker_num(), PSContext::instance()->initial_server_num(), - PSContext::instance()->scheduler_host(), PSContext::instance()->scheduler_port()); server_num_ = PSContext::instance()->initial_server_num(); if (running_) { MS_LOG(INFO) << "'Worker is already running.";