pcm-coordinator/rpc/pb/pcmCore.proto

syntax = "proto3";

package pcmCore;
option go_package = "/pcmCore";

message SyncInfoReq {
  int64 participantId = 1;
  repeated HpcInfo HpcInfoList = 2;
  repeated CloudInfo CloudInfoList = 3;
  repeated AiInfo AiInfoList = 4;
  repeated VmInfo VmInfoList = 5;
}

message AiInfo {
  int64 participantId = 1;
  int64 taskId = 2;
  string project_id = 3;
  string name = 4;
  string status = 5;
  string startTime = 6;
  int64 runningTime = 7;
  string result = 8;
  string jobId = 9;
  string createTime = 10;
  string imageUrl = 11;
  string command = 12;
  string flavorId = 13;
  string subscriptionId = 14;
  string itemVersionId = 15;

}

message CloudInfo {
  int64 participant = 1;
  int64 taskId = 2;
  string apiVersion = 3;
  string kind = 4;
  string namespace = 5;
  string name = 6;
  string status = 7;
  string startTime = 8;
  int64 runningTime = 9;
  string result = 10;
  string yamlString = 11;
  int64 id = 12;
}

message VmInfo {
  int64 participantId = 1;
  int64 taskId = 2;
  string name = 3;
  string flavor_ref = 4;
  string image_ref = 5;
  string network_uuid = 6;
  string block_uuid = 7;
  string source_type = 8;
  bool delete_on_termination = 9;
  string state = 10;
}

message HpcInfo {
  int64 participantId = 1;
  int64 taskId = 2;
  string jobId = 3;
  string name = 4;
  string status = 5;
  string startTime = 6;
  int64 runningTime = 7;
  string result = 8;
  string workDir = 9;
  string wallTime = 10;
  string cmdScript = 11;
  string derivedEs = 12;
  string cluster = 13;
  string blockId = 14;
  uint32 allocNodes = 15;
  uint32 allocCpu = 16;
  string version = 17;
  string account = 18;
  uint32 exitCode = 19;
  uint32 assocId = 20;
  string appType = 21;
  string appName = 22;
  string queue = 23;
  string submitType = 24;
  string nNode = 25;
  string stdOutFile = 26;
  string stdErrFile = 27;
  string stdInput = 28;
  string environment = 29;
}

message SyncInfoResp {
  int64 code = 1;
  string msg = 2;
}

message InfoListReq {
  int64 participantId = 1;
}

message InfoListResp {
  repeated HpcInfo HpcInfoList = 1;
  repeated CloudInfo CloudInfoList = 2;
  repeated AiInfo AiInfoList = 3;
}

// pcm core services
service pcmCore {

  // SyncInfo Synchronous data information
  rpc SyncInfo(SyncInfoReq) returns (SyncInfoResp);

  // InfoList
  rpc InfoList(InfoListReq) returns (InfoListResp);
}

// participantTenant 租户信息
message ParticipantTenant {
  string tenantName = 1; // 租户名称
}

// 集群标签
message ParticipantLabel {
  int64 id = 1;     // id
  string key = 2;   // 标签名
  string value = 3; // 标签值
}

enum MessageStatus {
  FAIL = 0;
  SUCCESS = 1;
  UNKNOWN = 2;
}

message HealthCheckResp {
  MessageStatus messageStatus = 1;
  int64 code = 2;
  string msg = 3;
}

message ParticipantPhyResp {
  int64 code = 1;
  string msg = 2;
  int64 participantId = 3; // participant 唯一标识
}

// 集群静态信息返回
message ListParticipantPhyResp {
  int64 code = 1;
  string msg = 2;
  repeated ParticipantPhyReq ParticipantPhys = 3;
}

// participantPhy 静态信息
message ParticipantPhyReq {
  string name = 1;              // 名称
  string address = 2;              // 地址
  string networkType = 4;       // 集群网络类型
  string networkBandwidth = 5;  // 集群网络带宽
  string storageType = 6;       // 集群存储类型
  string storageSpace = 7;      // 集群存储空间
  string storageAvailSpace = 8; // 集群存储可用空间
  string storageBandwidth = 9;  // 集群存储带宽
  string type = 10; // 参与者类型:CLOUD-数算集群;AI-智算集群；HPC-超算集群
  int64 tenantId = 11;                      // 租户id
  string tenantName = 12;                   // 租户名称
  repeated NodePhyInfo nodeInfo = 13;       // 节点信息
  int64 participantId = 14;                 // participant id
  repeated ParticipantLabel labelInfo = 15; // 标签信息
  repeated QueuePhyInfo queueInfo = 16;  // 队列信息
  int64 id = 17;                            // id
  string MetricsUrl = 18; //监控url
  string RpcAddress = 19;
}

// NodePhyInfo 节点信息
message NodePhyInfo {
  int64 id = 1;
  string nodeName = 2;  // 节点名称
  string osName = 3;    // 系统名称
  string osVersion = 4; // 系统版本
  string archType = 5;  // 架构类型
  string archName = 6;  // 架构名称
  string archFreq = 7;  // 架构频率
}

// QueuePhyInfo 队列信息
message QueuePhyInfo {
  int64 id = 1;               // id
  string aclHosts = 2;       // 可用节点，多个节点用逗号隔开
  string queueId = 3;        // 队列名称
  string text = 4 ;          // 队列名称
  string queueName = 5;        //队列名称
  string queNodes = 6;         // 队列节点总数
  string queMinNodect = 7;     // 队列最小节点数
  string queMaxNgpus = 8;      // 队列最大GPU卡数
  string queMaxPpn = 9;        // 使用该队列作业最大CPU核心数
  string queChargeRate = 10;   // 费率
  string queMaxNcpus = 11;      // 用户最大可用核心数
  string queMaxNdcus = 12;      // 队列总DCU卡数
  string queMinNcpus = 13;      // 队列最小CPU核数
  string queFreeNodes = 14;    // 队列空闲节点数
  string queMaxNodect = 15;    // 队列作业最大节点数
  string queMaxGpuPN = 16;     // 队列单作业最大GPU卡数
  string queMaxWalltime = 17; // 队列最大运行时间
  string queMaxDcuPN = 18;     // 队列单作业最大DCU卡数
  string queNcpus = 19; //队列cpu数
  string queFreeNcpus = 20; //队列空闲cpu数
}

// ParticipantHeartbeatReq 心跳请求
message ParticipantHeartbeatReq {
  int64 participantId = 1; // participantId
  string address = 2;
}

// ParticipantAvailInfo Participant可用信息
message ParticipantAvailReq {
  int64 id = 1;                             // id
  int64 availStorageSpace = 2;              // 集群存储可用空间
  int64 userNum = 3;                        // 用户数量
  int64 pendingJobNum = 4;                  // 待处理作业数量
  int64 runningJobNum = 5;                  // 运行作业数量
  int64 participantId = 6;                  // 集群静态信息id
  repeated NodeAvailInfo nodeAvailInfo = 7; // 节点可用信息
}

// NodeAvailInfo 节点可用信息
message NodeAvailInfo {
  int64 id = 1;                  // id
  string nodeName = 2;           // 节点名称
  int64 cpuTotal = 3;            // cpu核数
  double cpuUsable = 4;          // cpu可用率
  int64 diskTotal = 5;           // 磁盘空间
  int64 diskAvail = 6;           // 磁盘可用空间
  int64 memTotal = 7;            // 内存总数
  int64 memAvail = 8;            // 内存可用数
  int64 gpuTotal = 9;            // gpu总数
  int64 gpuAvail = 10;           // gpu可用数
  int64 participantId = 11; // 集群动态信息id
}

// 集群可用信息
message ListParticipantAvailResp {
  int64 code = 1;
  string msg = 2;
  repeated ParticipantAvailReq ParticipantAvails = 3;
}

message ParticipantResp {
  int64 code = 1;
  string msg = 2;
}
message ParticipantServiceResp {
  int64 code = 1;
  string msg = 2;
  repeated ClientInfo data = 3;
}

message ClientInfo {
  string address = 1;         // @gotags: redis:"address"
  int64 participantId = 2;    // @gotags: redis:"participantId"
  string clientState = 3;     // @gotags: redis:"clientState"
  int64 lastHeartbeat = 4;    // @gotags: redis:"lastHeartbeat"
}

message TenantInfo {
  int64 id = 1;
  string tenantName = 2;
  string tenantDesc = 3;
}

message TenantResp {
  int64 code = 1;
  string msg = 2;
  int64 id = 3;
}

message ListTenantResp {
  int64 code = 1;
  string msg = 2;
  repeated TenantInfo tenantInfos = 3;
}

// participant 参与者
service participantService {

  // registerParticipant Participant注册接口
  rpc registerParticipant(ParticipantPhyReq) returns (ParticipantPhyResp) {};

  // reportHeartbeat 心跳请求
  rpc reportHeartbeat(ParticipantHeartbeatReq) returns (HealthCheckResp) {};

  // reportAvailable 监控数据上报
  rpc reportAvailable(ParticipantAvailReq) returns (ParticipantResp) {}

  // listParticipant 服务列表
  rpc listParticipant(ParticipantTenant) returns (ParticipantServiceResp) {}

  // listAvailable 集群动态信息列表
  rpc listPhyAvailable(ParticipantTenant) returns (ListParticipantAvailResp) {}

  // listPhyInformation 集群静态信息列表
  rpc listPhyInformation(ParticipantTenant) returns (ListParticipantPhyResp) {};

  // registerTenant 注册租户信息
  rpc registerTenant(TenantInfo) returns (TenantResp) {};

  // listTenant 租户列表信息
  rpc listTenant(TenantInfo) returns (ListTenantResp) {};
}