add nodemanager v2

gen/proto/go/nodemanager/v2/service.pb.go

+// Code generated by protoc-gen-gogo. DO NOT EDIT.
+// source: nodemanager/v2/service.proto
+
+package nodemanagerv2
+
+import (
+	fmt "fmt"
+	proto "github.com/gogo/protobuf/proto"
+	math "math"
+)
+
+// Reference imports to suppress errors if they are not otherwise used.
+var _ = proto.Marshal
+var _ = fmt.Errorf
+var _ = math.Inf
+
+// This is a compile-time assertion to ensure that this generated file
+// is compatible with the proto package it is being compiled against.
+// A compilation error at this line likely means your copy of the
+// proto package needs to be updated.
+const _ = proto.GoGoProtoPackageIsVersion3 // please upgrade the proto package
+
+func init() { proto.RegisterFile("nodemanager/v2/service.proto", fileDescriptor_2b1eb586465a8941) }
+
+var fileDescriptor_2b1eb586465a8941 = []byte{
+	// 316 bytes of a gzipped FileDescriptorProto
+	0x1f, 0x8b, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0xff, 0x7c, 0x91, 0xc1, 0x4a, 0xc3, 0x40,
+	0x10, 0x86, 0x9b, 0x08, 0x1e, 0xd6, 0x52, 0x61, 0x8f, 0x41, 0x3d, 0x54, 0x05, 0x2f, 0x66, 0x65,
+	0xbd, 0xd5, 0x5b, 0x11, 0xbc, 0xd8, 0x5a, 0xaa, 0xd6, 0xa0, 0x05, 0xd9, 0x26, 0xc3, 0x36, 0xd4,
+	0x64, 0xe3, 0xce, 0x26, 0xe0, 0xeb, 0xf4, 0xe8, 0xa3, 0x88, 0x4f, 0xe1, 0xd1, 0xa7, 0x90, 0x26,
+	0xa9, 0x24, 0x11, 0x7b, 0xdb, 0xf9, 0xbf, 0x7f, 0xfe, 0xd9, 0x9d, 0x25, 0x7b, 0xb1, 0x0a, 0x20,
+	0x12, 0xb1, 0x90, 0xa0, 0x59, 0xc6, 0x19, 0x82, 0xce, 0x42, 0x1f, 0xdc, 0x44, 0x2b, 0xa3, 0x68,
+	0xa7, 0x42, 0xdd, 0x8c, 0x3b, 0xc7, 0x0d, 0xb7, 0x86, 0xd7, 0x14, 0xd0, 0x3c, 0x6b, 0xc0, 0x44,
+	0xc5, 0x58, 0xb6, 0xf1, 0xa5, 0x4d, 0xe8, 0x50, 0x05, 0x30, 0x28, 0x9c, 0xb7, 0x45, 0x26, 0xbd,
+	0x27, 0x9d, 0x31, 0xc8, 0x10, 0x0d, 0xe8, 0x07, 0xa5, 0x17, 0xa0, 0xe9, 0xbe, 0x5b, 0x1f, 0xe0,
+	0x16, 0xfa, 0x00, 0x10, 0x85, 0x04, 0xe7, 0xa0, 0x89, 0xcb, 0xc0, 0x92, 0x77, 0x5b, 0x27, 0xd6,
+	0x99, 0x45, 0x3d, 0xb2, 0x53, 0xea, 0xd7, 0x21, 0x1a, 0xda, 0xfd, 0xa7, 0x69, 0x05, 0xc7, 0xc5,
+	0x7d, 0x9d, 0xc3, 0x8d, 0x9e, 0xe2, 0x2d, 0xdd, 0x16, 0x7d, 0x22, 0xed, 0xcb, 0x10, 0x13, 0x61,
+	0xfc, 0xf9, 0x9d, 0xc0, 0x05, 0xfd, 0xd3, 0x56, 0xa5, 0xeb, 0xec, 0xa3, 0xcd, 0xa6, 0x75, 0x78,
+	0xff, 0xd3, 0x22, 0xd4, 0x57, 0x51, 0xc3, 0xdd, 0x6f, 0x97, 0xdb, 0x1a, 0xad, 0x36, 0x39, 0xb2,
+	0x1e, 0x6f, 0x64, 0x68, 0xe6, 0xe9, 0xcc, 0xf5, 0x55, 0xc4, 0x54, 0xf0, 0x86, 0x08, 0x29, 0xfe,
+	0x1e, 0x4e, 0xf3, 0x7d, 0xfb, 0xea, 0x85, 0x49, 0x88, 0x59, 0x5e, 0x30, 0xa9, 0x58, 0xfd, 0x97,
+	0x2e, 0x2a, 0x65, 0xc6, 0x97, 0xf6, 0xd6, 0xd0, 0xf3, 0xde, 0xed, 0xce, 0xb0, 0x32, 0x77, 0xc2,
+	0x3f, 0x6a, 0xc2, 0x74, 0xc2, 0xbf, 0x6c, 0xa7, 0x2e, 0x4c, 0xaf, 0x46, 0xfd, 0x01, 0x18, 0x11,
+	0x08, 0x23, 0xbe, 0xed, 0xdd, 0x0a, 0xec, 0xf5, 0x26, 0x7c, 0xb6, 0x9d, 0x4f, 0x3f, 0xff, 0x09,
+	0x00, 0x00, 0xff, 0xff, 0xd4, 0xe7, 0xd3, 0x21, 0x51, 0x02, 0x00, 0x00,
+}

gen/proto/go/nodemanager/v2/service_grpc.pb.go

+// Code generated by protoc-gen-go-grpc. DO NOT EDIT.
+// versions:
+// - protoc-gen-go-grpc v1.2.0
+// - protoc             (unknown)
+// source: nodemanager/v2/service.proto
+
+package nodemanagerv2
+
+import (
+	context "context"
+	grpc "google.golang.org/grpc"
+	codes "google.golang.org/grpc/codes"
+	status "google.golang.org/grpc/status"
+)
+
+// This is a compile-time assertion to ensure that this generated file
+// is compatible with the grpc package it is being compiled against.
+// Requires gRPC-Go v1.32.0 or later.
+const _ = grpc.SupportPackageIsVersion7
+
+// NodeManagerServiceClient is the client API for NodeManagerService service.
+//
+// For semantics around ctx use and closing/ending streaming RPCs, please refer to https://pkg.go.dev/google.golang.org/grpc/?tab=doc#ClientConn.NewStream.
+type NodeManagerServiceClient interface {
+	RegisterWorker(ctx context.Context, opts ...grpc.CallOption) (NodeManagerService_RegisterWorkerClient, error)
+	ManagerList(ctx context.Context, in *ManagerListRequest, opts ...grpc.CallOption) (*ManagerListResponse, error)
+	DispatchTask(ctx context.Context, in *DispatchTaskRequest, opts ...grpc.CallOption) (*DispatchTaskResponse, error)
+}
+
+type nodeManagerServiceClient struct {
+	cc grpc.ClientConnInterface
+}
+
+func NewNodeManagerServiceClient(cc grpc.ClientConnInterface) NodeManagerServiceClient {
+	return &nodeManagerServiceClient{cc}
+}
+
+func (c *nodeManagerServiceClient) RegisterWorker(ctx context.Context, opts ...grpc.CallOption) (NodeManagerService_RegisterWorkerClient, error) {
+	stream, err := c.cc.NewStream(ctx, &NodeManagerService_ServiceDesc.Streams[0], "/nodemanager.v2.NodeManagerService/RegisterWorker", opts...)
+	if err != nil {
+		return nil, err
+	}
+	x := &nodeManagerServiceRegisterWorkerClient{stream}
+	return x, nil
+}
+
+type NodeManagerService_RegisterWorkerClient interface {
+	Send(*WorkerMessage) error
+	Recv() (*ManagerMessage, error)
+	grpc.ClientStream
+}
+
+type nodeManagerServiceRegisterWorkerClient struct {
+	grpc.ClientStream
+}
+
+func (x *nodeManagerServiceRegisterWorkerClient) Send(m *WorkerMessage) error {
+	return x.ClientStream.SendMsg(m)
+}
+
+func (x *nodeManagerServiceRegisterWorkerClient) Recv() (*ManagerMessage, error) {
+	m := new(ManagerMessage)
+	if err := x.ClientStream.RecvMsg(m); err != nil {
+		return nil, err
+	}
+	return m, nil
+}
+
+func (c *nodeManagerServiceClient) ManagerList(ctx context.Context, in *ManagerListRequest, opts ...grpc.CallOption) (*ManagerListResponse, error) {
+	out := new(ManagerListResponse)
+	err := c.cc.Invoke(ctx, "/nodemanager.v2.NodeManagerService/ManagerList", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *nodeManagerServiceClient) DispatchTask(ctx context.Context, in *DispatchTaskRequest, opts ...grpc.CallOption) (*DispatchTaskResponse, error) {
+	out := new(DispatchTaskResponse)
+	err := c.cc.Invoke(ctx, "/nodemanager.v2.NodeManagerService/DispatchTask", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+// NodeManagerServiceServer is the server API for NodeManagerService service.
+// All implementations must embed UnimplementedNodeManagerServiceServer
+// for forward compatibility
+type NodeManagerServiceServer interface {
+	RegisterWorker(NodeManagerService_RegisterWorkerServer) error
+	ManagerList(context.Context, *ManagerListRequest) (*ManagerListResponse, error)
+	DispatchTask(context.Context, *DispatchTaskRequest) (*DispatchTaskResponse, error)
+	mustEmbedUnimplementedNodeManagerServiceServer()
+}
+
+// UnimplementedNodeManagerServiceServer must be embedded to have forward compatible implementations.
+type UnimplementedNodeManagerServiceServer struct {
+}
+
+func (UnimplementedNodeManagerServiceServer) RegisterWorker(NodeManagerService_RegisterWorkerServer) error {
+	return status.Errorf(codes.Unimplemented, "method RegisterWorker not implemented")
+}
+func (UnimplementedNodeManagerServiceServer) ManagerList(context.Context, *ManagerListRequest) (*ManagerListResponse, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method ManagerList not implemented")
+}
+func (UnimplementedNodeManagerServiceServer) DispatchTask(context.Context, *DispatchTaskRequest) (*DispatchTaskResponse, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method DispatchTask not implemented")
+}
+func (UnimplementedNodeManagerServiceServer) mustEmbedUnimplementedNodeManagerServiceServer() {}
+
+// UnsafeNodeManagerServiceServer may be embedded to opt out of forward compatibility for this service.
+// Use of this interface is not recommended, as added methods to NodeManagerServiceServer will
+// result in compilation errors.
+type UnsafeNodeManagerServiceServer interface {
+	mustEmbedUnimplementedNodeManagerServiceServer()
+}
+
+func RegisterNodeManagerServiceServer(s grpc.ServiceRegistrar, srv NodeManagerServiceServer) {
+	s.RegisterService(&NodeManagerService_ServiceDesc, srv)
+}
+
+func _NodeManagerService_RegisterWorker_Handler(srv interface{}, stream grpc.ServerStream) error {
+	return srv.(NodeManagerServiceServer).RegisterWorker(&nodeManagerServiceRegisterWorkerServer{stream})
+}
+
+type NodeManagerService_RegisterWorkerServer interface {
+	Send(*ManagerMessage) error
+	Recv() (*WorkerMessage, error)
+	grpc.ServerStream
+}
+
+type nodeManagerServiceRegisterWorkerServer struct {
+	grpc.ServerStream
+}
+
+func (x *nodeManagerServiceRegisterWorkerServer) Send(m *ManagerMessage) error {
+	return x.ServerStream.SendMsg(m)
+}
+
+func (x *nodeManagerServiceRegisterWorkerServer) Recv() (*WorkerMessage, error) {
+	m := new(WorkerMessage)
+	if err := x.ServerStream.RecvMsg(m); err != nil {
+		return nil, err
+	}
+	return m, nil
+}
+
+func _NodeManagerService_ManagerList_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(ManagerListRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(NodeManagerServiceServer).ManagerList(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/nodemanager.v2.NodeManagerService/ManagerList",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(NodeManagerServiceServer).ManagerList(ctx, req.(*ManagerListRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _NodeManagerService_DispatchTask_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(DispatchTaskRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(NodeManagerServiceServer).DispatchTask(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/nodemanager.v2.NodeManagerService/DispatchTask",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(NodeManagerServiceServer).DispatchTask(ctx, req.(*DispatchTaskRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+// NodeManagerService_ServiceDesc is the grpc.ServiceDesc for NodeManagerService service.
+// It's only intended for direct use with grpc.RegisterService,
+// and not to be introspected or modified (even as a copy)
+var NodeManagerService_ServiceDesc = grpc.ServiceDesc{
+	ServiceName: "nodemanager.v2.NodeManagerService",
+	HandlerType: (*NodeManagerServiceServer)(nil),
+	Methods: []grpc.MethodDesc{
+		{
+			MethodName: "ManagerList",
+			Handler:    _NodeManagerService_ManagerList_Handler,
+		},
+		{
+			MethodName: "DispatchTask",
+			Handler:    _NodeManagerService_DispatchTask_Handler,
+		},
+	},
+	Streams: []grpc.StreamDesc{
+		{
+			StreamName:    "RegisterWorker",
+			Handler:       _NodeManagerService_RegisterWorker_Handler,
+			ServerStreams: true,
+			ClientStreams: true,
+		},
+	},
+	Metadata: "nodemanager/v2/service.proto",
+}

nodemanager/nodemanager/v2/request_response.proto

+syntax = "proto3";
+
+package nodemanager.v2;
+import "base/gogo.proto";
+import "base/v1/odysseus.proto";
+import "nodemanager/v2/resource.proto";
+
+message WorkerMessage {
+  oneof message {
+    HeartbeatResponse heartbeat_response = 1;
+    GoodbyeMessage goodbye_message = 2;
+    // node info
+    NodeInfoResponse node_info = 3;
+    RegisteMessage registe_message = 4;
+    
+    
+    // task
+    SubmitTaskResult submit_task_result = 5;
+    FetchStandardTask fetch_standard_task = 6;
+    SubmitTaskAck submit_task_ack = 7;
+
+    // hardware info
+    DeviceInfoMessage device_info = 8;
+    DeviceUsageResponse device_usage = 9;
+    GPUUsageResponse gpu_usage = 10;
+
+    // model info
+    ModelsInfo models_info = 11;
+    AddModelInstalled add_model_installed = 12; // model is add to installed
+    DelModelInstalled del_model_installed = 13; // model is remove in installed-list.
+    AddModelRunning add_model_running = 14; // model is add to running
+    DelModelRunning del_mode_running = 15; // model is remove in running-list.
+
+    InstalledModelStatus installed_model_status = 16; // update installed model status
+    RunningModelStatus running_model_status = 17; // update running model status.
+  }
+}
+
+message AddModelInstalled {
+  repeated InstalledModel models = 1;
+}
+
+message DelModelInstalled {
+  repeated string model_ids = 1;
+}
+
+message AddModelRunning {
+  repeated RunningModel models = 1;
+}
+
+message DelModelRunning {
+  repeated string model_ids = 1;
+}
+
+message ManagerMessage {
+  oneof message {
+    HeartbeatRequest heartbeat_request = 1;
+    NodeInfoRequest node_info_request = 2;
+    GoodbyeMessage goodbye_message = 3;
+    // task 
+    PushTaskMessage push_task = 4;
+
+    // model image
+    ModelOperateRequest model_operate_request = 5;
+    ModelListRequest model_list_request = 9; // response is ModelsInfo
+
+    // hardware
+    DeviceInfoRequest device_info_request = 10; // response is HardwareInfo
+    DeviceUsageRequest device_usage_request = 11; // response is DeviceUsageResponse
+    GPUUsageRequest gpu_usage_request = 12; // response is GPUUsageResponse
+  }
+}
+
+
+message HeartbeatRequest {
+  uint64 timestamp = 1;
+}
+
+message HeartbeatResponse {
+  uint64 timestamp = 1;
+}
+
+message GoodbyeMessage {
+  string reason = 1;
+}
+
+message NodeInfoRequest {
+}
+
+message NodeInfoResponse {
+  NodeInfo info = 1;
+  HardwareInfo hardware = 2;
+  ModelsInfo models = 3;
+}
+
+
+message PushTaskMessage {
+  string task_id = 1;
+  base.v1.TaskKind task_kind = 2;
+  uint64 workload = 3;
+  string task_cmd = 4; // 
+  bytes task_param = 5;
+  uint64 task_type = 6;
+}
+
+message ModelOperateRequest {
+  // nodemanager 通过此消息通知 worker 安装或者删除模型
+  repeated ModelOperate model_operates = 1;
+}
+
+
+message ModelListRequest {
+}
+
+// message ModelStatusUpdate {
+//   repeated ModelStatus model_status = 1;
+// }
+
+message DeviceUsageRequest {
+}
+
+message DeviceUsageResponse {
+  HardwareUsage usage = 1;
+}
+
+message GPUUsageRequest {
+}
+
+message GPUUsageResponse {
+  repeated GPUUsage usages = 1;
+}
+
+message DeviceInfoRequest {
+}
+
+message DeviceInfoMessage {
+  HardwareInfo hardware = 1;
+  bytes device_signature = 2; // 由测试程序对设备信息进行签名
+}
+
+message RegisteMessage {
+  NodeInfo info = 1;
+  HardwareInfo hardware = 2;
+  ModelsInfo models = 3;
+  int64 timestamp = 4;
+  bytes device_signature = 5;
+}
+
+// container_signature = sign(hash(task_id+hash(task_param)+hash(task_result)))
+// miner_signature     = sign(hash(task_id+hash(task_param)+hash(task_result)))
+// manager_signature   = sign(hash(task_id+hash(task_param)+hash(task_result)+(reward_address).Bytes()+(miner_address).Bytes()+container_signature+miner_signature+workload))
+
+message SubmitTaskResult {
+  string task_id = 1;
+  bytes container_signature = 2; // 容器签名
+  bytes miner_signature = 3; // worker 签名
+  bytes task_result_header = 4; // result headyer
+  bytes task_result_body = 5; // result body.
+  bool is_successed = 6; // 任务是否成功
+  uint64 task_execute_duration = 7; // 任务实际执行时间, us 为单位
+  int32 task_execute_code = 8; // 任务执行结果状态码
+  string task_execute_error = 9; // 任务执行错误信息
+}
+
+message SubmitTaskAck {
+  string task_id = 1;
+  bool can_execute = 2;
+  int64 boot_up_time = 3;
+  int64 queue_wait_time = 4;
+  int64 execute_time = 5;
+}
+
+message FetchStandardTask {
+  uint64 task_type = 1; // fetch a standard task with special task_type.
+}
+
+// message for scheduler
+
+message DispatchTaskRequest {
+  base.v1.TaskContent task_data = 1;
+  string miner = 2;
+}
+
+message DispatchTaskResponse {
+  string task_id = 1;
+  string miner = 2;
+}
+
+message ManagerListRequest {
+  
+}
+
+message ManagerListResponse {
+  repeated NodeManagerInfo managers = 1 ;
+}
\ No newline at end of file

nodemanager/nodemanager/v2/resource.proto

+syntax = "proto3";
+
+package nodemanager.v2;
+
+message NodeManagerInfo {
+    string publickey = 1;
+    string endpoint = 2;
+}
+
+message NodeInfo {
+    string miner_pubkey = 1;
+    string benefit_address = 2;
+}
+
+enum ModelOperateType {
+    INSTALL = 0;
+    DELETE = 1;
+    RUN = 2;
+    STOP = 3;
+}
+
+enum ModelStatusType {
+    NOTEXIST = 0;
+    INSTALLED = 1;
+    DELETED = 2;
+    RUNNING = 3;
+    STOPPED = 4;
+}
+
+
+message ModelOperate {
+    string model_id = 1;
+    ModelOperateType operate = 2;
+}
+
+message ModelStatus {
+    string model_id = 1;
+    ModelStatusType status = 2;
+}
+  
+
+message ModelsInfo {
+    repeated InstalledModel installed_models = 1;
+    repeated RunningModel running_models = 2;
+}
+
+message InstalledModel {
+    string model_id = 1;
+    int64 disk_size = 2;
+    int64 installed_time = 3;
+    int64 last_run_time = 4;
+}
+
+message InstalledModelStatus {
+    string model_id = 1;
+    int64 last_run_time = 2;
+}
+
+message RunningModel {
+    string model_id = 1;
+    int32 gpu_seq = 2;
+    int64 gpu_ram = 3;
+    int64 started_time = 4;
+    int64 last_work_time = 5;
+    int32 total_run_count = 6;
+    int32 wait_time = 7;
+}
+
+message RunningModelStatus {
+    string model_id = 1;
+    int64 last_work_time = 2;
+    int32 total_run_count = 3;
+    int32 wait_time = 4;
+}
+
+message HardwareInfo {
+    CPUInfo CPU = 1;
+    repeated GPUInfo GPU = 2;
+    MemoryInfo RAM = 3;
+    DiskInfo DISK = 4;
+    NetInfo NET = 5;
+}
+
+message HardwareUsage {
+    int32 cpu_usage = 1;
+    int32 ram_usage = 2;
+    int32 disk_usage = 3;
+    int32 net_bandwidth = 4;
+}
+
+message GPUUsage {
+    int32 seq = 1;
+    int64 mem_free = 2;
+    int32 usage = 3;
+    int32 temp = 4;
+    int32 power_rt = 5;
+}
+
+message CPUInfo {
+    string model = 1;
+    int32 number = 2;
+    int32 cores = 3;
+    int32 threads = 4;
+    int32 usage = 5; // 使用率-百分比
+}
+
+message GPUInfo {
+    int32 seq = 1;
+    string uuid = 2;
+    string model = 3;
+    int32 performance = 4; // 算力
+    int32 power_rating = 5; // 额定功率
+    int64 mem_total = 6; // 显存总量-字节
+    int64 mem_free = 7; // 显存剩余量-字节
+    int32 usage = 8; // gpu 使用率-百分比
+    int32 temp = 9;  // 温度
+    int32 power_rt = 10; // 实时功率
+}
+
+message MemoryInfo {
+    int64 total = 1;
+    int64 free = 2;
+}
+
+message DiskInfo {
+    int64 total = 1;
+    int64 free = 2;
+}
+
+message NetInfo {
+    string ip = 1;
+    string mac = 2;
+    int32 bandwidth = 3;
+}
\ No newline at end of file

nodemanager/nodemanager/v2/service.proto

+syntax = "proto3";
+
+package nodemanager.v2;
+import "nodemanager/v2/request_response.proto";
+
+service NodeManagerService {
+  rpc RegisterWorker(stream WorkerMessage) returns (stream ManagerMessage) {}
+
+  rpc ManagerList(ManagerListRequest) returns (ManagerListResponse) {}
+
+  rpc DispatchTask(DispatchTaskRequest) returns (DispatchTaskResponse) {}
+}