update msg resp

cmd/rootcmd.go

@@ -12,6 +12,7 @@ import (
 	"github.com/spf13/cobra"
 	"github.com/spf13/viper"
 	"os"
+	"strings"
 )
 
 var (
@@ -20,8 +21,8 @@ var (
 )
 
 func init() {
-	RootCmd.PersistentFlags().StringVarP(&rewardAddr, "reward", "r", "0x40EC4256fcBCA69CdbAc942594caeC79FBE10494", "please enter a reward address")
-	RootCmd.PersistentFlags().StringVarP(&externalIp, "externalIp", "e", "192.168.1.120", "please enter server external ip address")
+	RootCmd.PersistentFlags().StringVarP(&rewardAddr, "reward", "r", "", "please enter a reward address")
+	RootCmd.PersistentFlags().StringVarP(&externalIp, "externalIp", "e", "", "please enter server external ip address")
 	RootCmd.PersistentFlags().StringVarP(&opSys, "opSys", "s", "", "please enter you op sys name : win、linux")
 	RootCmd.PersistentFlags().BoolVarP(&debug, "debug", "d", false, "set log level debug")
 	cobra.OnInitialize(initConfig)
@@ -58,7 +59,7 @@ var RootCmd = &cobra.Command{
 			}
 			isExist := false
 			for _, acc := range nm.HistoryBenefitAcc {
-				if acc.Address == rewardAddr {
+				if strings.ToLower(acc.Address) == strings.ToLower(rewardAddr) {
 					isExist = true
 				}
 			}
@@ -70,7 +71,7 @@ var RootCmd = &cobra.Command{
 					return
 				}
 			}
-			//go nm.StartMonitor()
+			go nm.StartMonitor()
 		}
 		beego.Run()
 	},

conf/config.go

@@ -18,7 +18,8 @@ type Config struct {
 	ExternalIp               string
 	SignPublicAddress        common.Address
 	SignPrivateKey           *ecdsa.PrivateKey
-	RunMode                  int    `json:"init_run_mode"`
+	HardwareUrl              string `json:"hardware_url" mapstructure:"hardware_url"`
+	RunMode                  int    `json:"init_run_mode"  mapstructure:"init_run_mode"`
 	NmSeed                   string `json:"nm_seed" mapstructure:"nm_seed"`
 	HeartRespTimeSecond      int64  `json:"heart_response" mapstructure:"heart_response"`
 	NodeManagerNum           int64  `json:"node_manager_num" mapstructure:"node_manager_num"`

config.json

 {
-  "nm_seed": "43.198.29.144:10001",
+  "nm_seed": "52.221.177.10:10001",
   "api_url": "https://aigic.ai/admin/api/task/taskheat",
   "node_manager_num": 1,
   "heart_response": 60,
@@ -12,5 +12,6 @@
   "replicate_image_name_suffix": "docker.aigic.ai/ai",
   "is_stop_last_container": true,
   "disk_usage":80,
-  "init_run_mode": 1
+  "init_run_mode": 1,
+  "hardware_url": "http://47.94.59.74:8005/hw"
 }
\ No newline at end of file

controllers/NodeController.go

@@ -9,6 +9,7 @@ import (
 	"example.com/m/utils"
 	nodemanagerV2 "github.com/odysseus/odysseus-protocol/gen/proto/go/nodemanager/v2"
 	"io"
+	"strings"
 	"time"
 )
 
@@ -54,7 +55,7 @@ func (c *NodeController) SetBenefitAddress() {
 	}
 	isExist := false
 	for _, s := range nm.HistoryBenefitAcc {
-		if s.Address == req.Address && !s.IsDel {
+		if strings.ToLower(s.Address) == strings.ToLower(req.Address) && !s.IsDel {
 			isExist = true
 		}
 	}
@@ -207,7 +208,7 @@ func (c *NodeController) DelBenefitAddress() {
 	}
 	isExist := false
 	for _, s := range nm.HistoryBenefitAcc {
-		if s.Address == req.Address {
+		if strings.ToLower(s.Address) == strings.ToLower(req.Address) {
 			s.IsDel = true
 			isExist = true
 		}

controllers/StateController.go

@@ -27,7 +27,7 @@ func (c *StateController) GetRunningState() {
 }
 
 func (c *StateController) GetRunningTp() {
-	info := utils.GetHardwareInfo()
+	info := utils.GetHardwareInfo(conf.GetConfig().HardwareUrl)
 	if info == nil {
 		c.ResponseInfo(500, "get running tp failed", 0)
 		return
@@ -69,7 +69,7 @@ func (c *StateController) GetWorkerInfo() {
 }
 
 func (c *StateController) GetListGpuInfo() {
-	info := utils.GetHardwareInfo()
+	info := utils.GetHardwareInfo(conf.GetConfig().HardwareUrl)
 	if info != nil && info.Data != nil {
 		c.ResponseInfo(200, "get list gpu info successful", info.Data.Gpus)
 		return
@@ -89,7 +89,7 @@ func (c *StateController) GetGpuUsageInfo() {
 		c.ResponseInfo(500, "param error", "")
 		return
 	}
-	info := utils.GetHardwareInfo()
+	info := utils.GetHardwareInfo(conf.GetConfig().HardwareUrl)
 	if info != nil {
 		for _, gpu := range info.Data.Gpus {
 			if gpu.Seq == req.Seq {
@@ -102,7 +102,7 @@ func (c *StateController) GetGpuUsageInfo() {
 }
 
 func (c *StateController) GetOtherHardwareInfo() {
-	info := utils.GetHardwareInfo()
+	info := utils.GetHardwareInfo(conf.GetConfig().HardwareUrl)
 	var diskTotal, diskFree int64
 	for _, disk := range info.Data.Disk {
 		for _, point := range disk.MountPoints {

largeModel/model_handler.go

@@ -58,6 +58,11 @@ func (m *ModelHandler) MonitorModelInfo() {
 				log.Warn("Response data is empty")
 				continue
 			}
+			imageMap, err := m.dockerOp.PsImageNameMap()
+			if err != nil {
+				log.Error("Error getting image name map from client failed:", err)
+				continue
+			}
 			modelInfosResp := resp.Data
 			for _, modelInfo := range modelInfosResp {
 				if modelInfo.ImageName == "" {
@@ -68,7 +73,7 @@ func (m *ModelHandler) MonitorModelInfo() {
 				if len(split) != 2 {
 					continue
 				}
-				log.WithField("name", modelInfo.ImageName).Info("The image name is already")
+				log.WithField("name", modelInfo.ImageName).Info("The image name is")
 				m.dockerOp.SignApi[modelInfo.ImageName] = modelInfo.SignUrl
 				model, _ := db.GetModel(modelInfo.ImageName)
 				if model != nil {
@@ -78,14 +83,16 @@ func (m *ModelHandler) MonitorModelInfo() {
 						log.WithError(err).Error("Put db error")
 						continue
 					}
+					log.WithField("name", modelInfo.ImageName).Info("The image updated")
 				} else {
 					err := db.PutModel(modelInfo.ImageName, modelInfo)
 					if err != nil {
 						log.WithError(err).Error("Put db error")
 						continue
 					}
+					log.WithField("name", modelInfo.ImageName).Info("The image add")
 				}
-				if modelInfo.PublishStatus == models.ModelPublishStatusYes {
+				if !imageMap[modelInfo.ImageName] && modelInfo.PublishStatus == models.ModelPublishStatusYes {
 					log.WithField("model image name", modelInfo.ImageName).Info("pulling image")
 					go m.dockerOp.PullImage(model.ImageName)
 				}
@@ -180,34 +187,3 @@ func (m *ModelHandler) MonitorModelStatus() {
 		}
 	}
 }
-
-func (m *ModelHandler) isResourceEnough(modelInfo *models.ModelInfo) bool {
-	return true
-}
-
-func (m *ModelHandler) checkGpuUsage(modelInfo *models.ModelInfo) bool {
-	return false
-}
-
-func (m *ModelHandler) checkDiskUsage(modelInfo *models.ModelInfo) bool {
-	totalSize, usedSize, availSize, usageSize, err := m.dockerOp.GetDockerInfo()
-	if err != nil {
-		log.WithError(err).Error("Disk resource is not enough")
-		return false
-	}
-	log.WithField("TotalSize:", totalSize).WithField("UsedSize:", usedSize).WithField("AvailSize:", availSize).Info("The disk info")
-	if conf.GetConfig().DiskUsage < usageSize {
-		return false
-	}
-	modelDiskSizeGB, err := strconv.ParseInt(modelInfo.HardwareRequire.DiskSize, 10, 64)
-	if err != nil {
-		log.Errorf("Error parsing model disk usage failed: %v", err)
-		return false
-	}
-	modelDiskSizeByte := modelDiskSizeGB * 1024 * 1024
-	if availSize < modelDiskSizeByte {
-		log.Error("The hard drive is running out of space")
-		return false
-	}
-	return true
-}

models/node_manager.go

@@ -70,16 +70,16 @@ type GpuInfo struct {
 }
 
 type Gpu struct {
-	Seq         int64  `json:"seq"`
+	Seq         int32  `json:"seq"`
 	Uuid        string `json:"uuid"`
 	Model       string `json:"model"`
-	Performance int64  `json:"performance"`
-	PowerRating int64  `json:"power_rating"`
+	Performance int32  `json:"performance"`
+	PowerRating int32  `json:"power_rating"`
 	MemTotal    int64  `json:"mem_total"`
 	MemFree     int64  `json:"mem_free"`
-	Usage       int64  `json:"usage"`
-	Temp        int64  `json:"temp"`
-	PowerRt     int64  `json:"power_rt"`
+	Usage       int32  `json:"usage"`
+	Temp        int32  `json:"temp"`
+	PowerRt     int32  `json:"power_rt"`
 }
 
 type Cpu struct {

models/req_resp.go

@@ -35,7 +35,7 @@ type WorkerAccount struct {
 }
 
 type GpuUsageReq struct {
-	Seq int64 `json:"seq"`
+	Seq int32 `json:"seq"`
 }
 
 type OtherHardwareInfoResp struct {

models/validator.go

-package models

nm/msg_resp.go

@@ -362,7 +362,7 @@ func DelModelRunningResp(params ...interface{}) *nodemanagerV2.WorkerMessage {
 }
 
 func getHardwareInfo() *nodemanagerV2.HardwareInfo {
-	hardwareInfo := utils.GetHardwareInfo()
+	hardwareInfo := utils.GetHardwareInfo(conf.GetConfig().HardwareUrl)
 	gpusInfo := make([]*nodemanagerV2.GPUInfo, 0)
 	var diskTotal, diskFree int64
 	for _, disk := range hardwareInfo.Data.Disk {
@@ -383,6 +383,21 @@ func getHardwareInfo() *nodemanagerV2.HardwareInfo {
 			bandWidth = net.Speed
 		}
 	}
+	for _, gpu := range hardwareInfo.Data.Gpus {
+		gpuInfo := &nodemanagerV2.GPUInfo{
+			Seq:         gpu.Seq,
+			Uuid:        gpu.Uuid,
+			Model:       gpu.Model,
+			Performance: gpu.Performance,
+			PowerRating: gpu.PowerRating,
+			MemTotal:    gpu.MemTotal,
+			MemFree:     gpu.MemFree,
+			Usage:       gpu.Usage,
+			Temp:        gpu.Temp,
+			PowerRt:     gpu.PowerRt,
+		}
+		gpusInfo = append(gpusInfo, gpuInfo)
+	}
 	res := &nodemanagerV2.HardwareInfo{
 		CPU: &nodemanagerV2.CPUInfo{
 			Model:   hardwareInfo.Data.Cpus.Model,

nm/start.go

@@ -34,9 +34,9 @@ func StartMonitor() {
 
 	monitorNm := NewMonitorNm(dockerOp, modelHandler)
 
-	go modelHandler.MonitorModelInfo()
+	//go modelHandler.MonitorModelInfo()
 
-	go modelHandler.MonitorModelStatus()
+	//go modelHandler.MonitorModelStatus()
 
 	go monitorNm.monitorNodeManagerSeed()
 

utils/util.go

@@ -310,8 +310,8 @@ func readAndDecryptFile(key []byte, filename string) ([]byte, error) {
 	return decryptedData, nil
 }
 
-func GetHardwareInfo() *models.HardwareInfoRep {
-	resp, err := http.Get("http://124.193.167.71:5000/hw")
+func GetHardwareInfo(url string) *models.HardwareInfoRep {
+	resp, err := http.Get(url)
 	if err != nil {
 		log.Error("Error creating request")
 		return nil