add PreparingSleep status

.gitignore

@@ -3,9 +3,6 @@ cmake-build-*
 build/
 .cache/
 
-.cache/*
-build/*
-
 third_party/*
 
 # Header only

dependencies/cmake/CMakeLists.txt

@@ -28,7 +28,6 @@ add_subdirectory(mongodb-cxx-driver)
 add_subdirectory(ranges-v3)
 add_subdirectory(backward-cpp)
 add_subdirectory(fpm)
-add_subdirectory(influxdb-cxx)
 add_subdirectory(fast-cpp-csv-parser)
 #add_subdirectory(mariadb-connector-c)
 

dependencies/cmake/fast-cpp-csv-parser/CMakeLists.txt

@@ -4,23 +4,22 @@ set(CMAKE_POLICY_DEFAULT_CMP0069 NEW)
 include(FetchContent)
 
 if (CRANE_USE_GITEE_SOURCE)
-    set(CSV_PARSER_URL "https://gitee.com/mirrors/fast-cpp-csv-parser/raw/master/csv.h")
+    # TODO: replace with gitee source
 else ()
     set(CSV_PARSER_URL "https://raw.githubusercontent.com/ben-strasser/fast-cpp-csv-parser/master/csv.h")
 endif ()
 
 FetchContent_Declare(
     fast_cpp_csv_parser
     URL ${CSV_PARSER_URL}
-    DOWNLOAD_NO_EXTRACT TRUE  # 因为是单个头文件
+    DOWNLOAD_NO_EXTRACT TRUE
     INACTIVITY_TIMEOUT 5
 )
 
 FetchContent_GetProperties(fast_cpp_csv_parser)
 if(NOT fast_cpp_csv_parser_POPULATED)
     FetchContent_Populate(fast_cpp_csv_parser)
 
-    # 创建 INTERFACE 库
     add_library(fast_cpp_csv_parser INTERFACE)
     target_include_directories(fast_cpp_csv_parser INTERFACE 
         ${fast_cpp_csv_parser_SOURCE_DIR}

etc/config.yaml

@@ -72,10 +72,24 @@ Nodes:
   - name: "cn[15-16]"
     cpu: 2
     memory: 2G
+    bmc:
+      ip: "10.10.10.10"
+      username: "ADMIN"
+      password: "ADMIN"
+    ssh:
+      username: "root"
+      password: "root"
 
   - name: "cn[17-18]"
     cpu: 2
     memory: 2G
+    bmc:
+      ip: "10.10.10.11"
+      username: "ADMIN"
+      password: "ADMIN"
+    ssh:
+      username: "root"
+      password: "root"
     gres:
       - name: gpu
         type: a100

etc/database.yaml

@@ -11,3 +11,10 @@ DbHost: localhost
 DbPort: 27017
 DbReplSetName: crane_rs
 DbName: crane_db
+
+# InfluxDB settings
+InfluxDbUrl: YOUR_INFLUXDB_URL
+InfluxDbToken: "YOUR_INFLUXDB_TOKEN"
+InfluxDbOrg: YOUR_INFLUXDB_ORG
+InfluxDbNodeBucket: YOUR_INFLUXDB_NODE_BUCKET
+InfluxDbTaskBucket: YOUR_INFLUXDB_TASK_BUCKET

protos/Crane.proto

@@ -607,6 +607,22 @@ message QueryCranedRemoteMetaReply {
   CranedRemoteMeta craned_remote_meta = 2;
 }
 
+message QueryCranedNICInfoRequest {}
+
+message QueryCranedNICInfoReply{
+  bool ok = 1;
+  string interface_name = 2;
+  string mac_address = 3;     
+  string error_message = 4; 
+}
+
+message SuspendCranedRequest {}
+
+message SuspendCranedReply {
+  bool ok = 1;
+  string error_message = 2;
+}
+
 message StreamCrunRequest {
   enum CrunRequestType {
     TASK_REQUEST = 0;
@@ -807,6 +823,10 @@ service CraneCtld {
   rpc QueryPartitionInfo(QueryPartitionInfoRequest) returns (QueryPartitionInfoReply);
   rpc ModifyTask(ModifyTaskRequest) returns (ModifyTaskReply);
   rpc ModifyNode(ModifyCranedStateRequest) returns (ModifyCranedStateReply);
+  rpc SleepCraned(SleepCranedRequest) returns (SleepCranedReply);
+  rpc WakeupCraned(WakeupCranedRequest) returns (WakeupCranedReply);
+  rpc ShutdownCraned(ShutdownCranedRequest) returns (ShutdownCranedReply);
+  rpc PowerOnCraned(PowerOnCranedRequest) returns (PowerOnCranedReply);
 
   /* RPCs called from cacctmgr */
   rpc AddAccount(AddAccountRequest) returns (AddAccountReply);
@@ -832,11 +852,6 @@ service CraneCtld {
 
   /* common RPCs */
   rpc QueryTasksInfo(QueryTasksInfoRequest) returns (QueryTasksInfoReply);
-
-  rpc SleepCraned(SleepCranedRequest) returns (SleepCranedReply) {}
-  rpc WakeupCraned(WakeupCranedRequest) returns (WakeupCranedReply) {}
-  rpc ShutdownCraned(ShutdownCranedRequest) returns (ShutdownCranedReply) {}
-  rpc PowerOnCraned(PowerOnCranedRequest) returns (PowerOnCranedReply) {}
 }
 
 service Craned {
@@ -849,7 +864,9 @@ service Craned {
   rpc ReleaseCgroupForTasks(ReleaseCgroupForTasksRequest) returns(ReleaseCgroupForTasksReply);
 
   rpc QueryCranedRemoteMeta(QueryCranedRemoteMetaRequest) returns(QueryCranedRemoteMetaReply);
-
+  rpc QueryCranedNICInfo(QueryCranedNICInfoRequest) returns (QueryCranedNICInfoReply);
+  rpc SuspendCraned(SuspendCranedRequest) returns (SuspendCranedReply);
+
   /*
   If the task is an interactive task, the resource uuid is also revoked.
    If there's no process in this interactive task, just deallocate all the resources.

protos/PublicDefs.proto

@@ -85,9 +85,10 @@ enum CranedResourceState {
   CRANE_MIX = 2;         
 
   // Not Running
-  CRANE_SLEEPED = 4;     
-  CRANE_SHUTDOWN = 5;    
-  CRANE_WAKING_UP = 6;   
+  CRANE_SLEEPED = 3;     
+  CRANE_SHUTDOWN = 4;    
+  CRANE_WAKING_UP = 5;   
+  CRANE_PREPARING_SLEEP = 6;
   CRANE_POWERING_UP = 7; 
   CRANE_SHUTTING_DOWN = 8;
   CRANE_UNKNOWN = 9;     

src/CraneCtld/CMakeLists.txt

@@ -59,7 +59,6 @@ target_link_libraries(cranectld PRIVATE
 
         Backward::Interface
 
-        influxdb_cpp
         CURL::libcurl 
         fast_cpp_csv_parser
         ssh

src/CraneCtld/CraneCtld.cpp

@@ -362,17 +362,14 @@ void ParseConfig(int argc, char** argv) {
 
           if (node["bmc"]) {
             auto bmc = node["bmc"].as<YAML::Node>();
+            // TODO: Communicate how to set the BMC IP address format.
             node_ptr->bmc.ip = bmc["ip"].as<std::string>();
-            node_ptr->bmc.port = bmc["port"].as<uint32_t>();
             node_ptr->bmc.username = bmc["username"].as<std::string>();
             node_ptr->bmc.password = bmc["password"].as<std::string>();
-            node_ptr->bmc.interface = bmc["interface"].as<std::string>();
           }
 
           if (node["ssh"]) {
             auto ssh = node["ssh"].as<YAML::Node>();
-            node_ptr->ssh.ip = ssh["ip"].as<std::string>();
-            node_ptr->ssh.port = ssh["port"].as<uint32_t>();
             node_ptr->ssh.username = ssh["username"].as<std::string>();
             node_ptr->ssh.password = ssh["password"].as<std::string>();
           }

src/CraneCtld/CranedKeeper.cpp

@@ -255,6 +255,46 @@ CraneErr CranedStub::QueryCranedRemoteMeta(CranedRemoteMeta *meta) {
   return CraneErr::kGenericFailure;
 }
 
+CraneErr CranedStub::QueryCranedNICInfo(CranedRemoteMeta *meta) {
+  using crane::grpc::QueryCranedNICInfoReply;
+  using crane::grpc::QueryCranedNICInfoRequest;
+  ClientContext context;
+  Status grpc_status;
+
+  QueryCranedNICInfoRequest request;
+  QueryCranedNICInfoReply reply;
+
+  auto status = m_stub_->QueryCranedNICInfo(&context, request, &reply);
+  if (!status.ok()) {
+    CRANE_ERROR("Failed to query NIC info for craned {}: {}", m_craned_id_,
+                status.error_message());
+    return CraneErr::kRpcFailure;
+  }
+
+  meta->nic.interface_name = reply.interface_name();
+  meta->nic.mac_address = reply.mac_address();
+  return CraneErr::kOk;
+}
+
+CraneErr CranedStub::SuspendCraned() {
+  using crane::grpc::SuspendCranedReply;
+  using crane::grpc::SuspendCranedRequest;
+
+  ClientContext context;
+  Status status;
+  SuspendCranedRequest request;
+  SuspendCranedReply reply;
+
+  status = m_stub_->SuspendCraned(&context, request, &reply);
+  if (!status.ok()) {
+    CRANE_ERROR("SuspendCraned RPC for Node {} returned with status not ok: {}",
+                m_craned_id_, status.error_message());
+    return CraneErr::kRpcFailure;
+  }
+
+  return CraneErr::kOk;
+}
+
 crane::grpc::ExecuteTasksRequest CranedStub::NewExecuteTasksRequests(
     const CranedId &craned_id, const std::vector<TaskInCtld *> &tasks) {
   crane::grpc::ExecuteTasksRequest request;
@@ -726,6 +766,7 @@ void CranedKeeper::ConnectCranedNode_(CranedId const &craned_id) {
   craned->m_stub_ = crane::grpc::Craned::NewStub(craned->m_channel_);
 
   craned->m_craned_id_ = craned_id;
+  craned->m_craned_ip_ = ip_addr;
   craned->m_clean_up_cb_ = CranedChannelConnectFail_;
 
   CqTag *tag = m_tag_sync_allocator_->new_object<CqTag>(

src/CraneCtld/CranedKeeper.h

@@ -60,8 +60,14 @@ class CranedStub {
 
   CraneErr QueryCranedRemoteMeta(CranedRemoteMeta *meta);
 
+  CraneErr QueryCranedNICInfo(CranedRemoteMeta *meta);
+
+  CraneErr SuspendCraned();
+
   bool Invalid() const { return m_invalid_; }
 
+  std::string GetCranedIp() const { return m_craned_ip_; }
+
  private:
   CranedKeeper *m_craned_keeper_;
 
@@ -77,6 +83,7 @@ class CranedStub {
   uint32_t m_failure_retry_times_;
 
   CranedId m_craned_id_;
+  std::string m_craned_ip_;
 
   // void* parameter is m_data_. Used to free m_data_ when CranedStub is being
   // destructed.