Merge branch 'V20211228' into fix-1095

models/cloudbrain.go

@@ -88,6 +88,9 @@ type Cloudbrain struct {
 	UpdatedUnix      timeutil.TimeStamp `xorm:"INDEX updated"`
 	Duration         int64
 	TrainJobDuration string
+	Image            string    //GPU镜像名称
+	GpuQueue         string    //GPU类型即GPU队列
+	ResourceSpecId   int       //GPU规格id
 	DeletedAt        time.Time `xorm:"deleted"`
 	CanDebug         bool      `xorm:"-"`
 	CanDel           bool      `xorm:"-"`
@@ -102,7 +105,7 @@ type Cloudbrain struct {
 	IsLatestVersion string //是否是最新版本，1是，0否
 	CommitID        string //提交的仓库代码id
 	PreVersionName  string //父版本名称
-	ComputeResource string //计算资源，例如npu
+	ComputeResource string `xorm:"-"`	//计算资源，例如npu
 	EngineID        int64  //引擎id
 
 	TrainUrl          string //输出的obs路径

modules/cloudbrain/cloudbrain.go

@@ -1,6 +1,8 @@
 package cloudbrain
 
 import (
+	"code.gitea.io/gitea/modules/storage"
+	"encoding/json"
 	"errors"
 	"strconv"
 
@@ -107,6 +109,9 @@ func GenerateTask(ctx *context.Context, jobName, image, command, uuid, codePath,
 		uuid
 
 	var resourceSpec *models.ResourceSpec
+	if ResourceSpecs == nil {
+		json.Unmarshal([]byte(setting.ResourceSpecs), &ResourceSpecs)
+	}
 	for _, spec := range ResourceSpecs.ResourceSpec {
 		if resourceSpecId == spec.Id {
 			resourceSpec = spec
@@ -185,28 +190,142 @@ func GenerateTask(ctx *context.Context, jobName, image, command, uuid, codePath,
 		},
 	})
 	if err != nil {
-		log.Error("CreateJob failed:", err.Error())
+		log.Error("CreateJob failed:", err.Error(), ctx.Data["MsgID"])
 		return err
 	}
 	if jobResult.Code != Success {
-		log.Error("CreateJob(%s) failed:%s", jobName, jobResult.Msg)
+		log.Error("CreateJob(%s) failed:%s", jobName, jobResult.Msg, ctx.Data["MsgID"])
 		return errors.New(jobResult.Msg)
 	}
 
 	var jobID = jobResult.Payload["jobId"].(string)
 	err = models.CreateCloudbrain(&models.Cloudbrain{
-		Status:      string(models.JobWaiting),
-		UserID:      ctx.User.ID,
-		RepoID:      ctx.Repo.Repository.ID,
-		JobID:       jobID,
-		JobName:     jobName,
-		SubTaskName: SubTaskName,
-		JobType:     jobType,
-		Type:        models.TypeCloudBrainOne,
-		Uuid:        uuid,
+		Status:         string(models.JobWaiting),
+		UserID:         ctx.User.ID,
+		RepoID:         ctx.Repo.Repository.ID,
+		JobID:          jobID,
+		JobName:        jobName,
+		SubTaskName:    SubTaskName,
+		JobType:        jobType,
+		Type:           models.TypeCloudBrainOne,
+		Uuid:           uuid,
+		Image:          image,
+		GpuQueue:       gpuQueue,
+		ResourceSpecId: resourceSpecId,
+	})
+
+	if err != nil {
+		return err
+	}
+
+	return nil
+}
+
+func RestartTask(ctx *context.Context, task *models.Cloudbrain) error {
+	dataActualPath := setting.Attachment.Minio.RealPath +
+		setting.Attachment.Minio.Bucket + "/" +
+		setting.Attachment.Minio.BasePath +
+		models.AttachmentRelativePath(task.Uuid) +
+		task.Uuid
+	jobName := task.JobName
+
+	var resourceSpec *models.ResourceSpec
+	if ResourceSpecs == nil {
+		json.Unmarshal([]byte(setting.ResourceSpecs), &ResourceSpecs)
+	}
+	for _, spec := range ResourceSpecs.ResourceSpec {
+		if task.ResourceSpecId == spec.Id {
+			resourceSpec = spec
+		}
+	}
+
+	if resourceSpec == nil {
+		log.Error("no such resourceSpecId(%d)", task.ResourceSpecId, ctx.Data["MsgID"])
+		return errors.New("no such resourceSpec")
+	}
+
+	jobResult, err := CreateJob(jobName, models.CreateJobParams{
+		JobName:    jobName,
+		RetryCount: 1,
+		GpuType:    task.GpuQueue,
+		Image:      task.Image,
+		TaskRoles: []models.TaskRole{
+			{
+				Name:                  SubTaskName,
+				TaskNumber:            1,
+				MinSucceededTaskCount: 1,
+				MinFailedTaskCount:    1,
+				CPUNumber:             resourceSpec.CpuNum,
+				GPUNumber:             resourceSpec.GpuNum,
+				MemoryMB:              resourceSpec.MemMiB,
+				ShmMB:                 resourceSpec.ShareMemMiB,
+				Command:               Command,
+				NeedIBDevice:          false,
+				IsMainRole:            false,
+				UseNNI:                false,
+			},
+		},
+		Volumes: []models.Volume{
+			{
+				HostPath: models.StHostPath{
+					Path:      storage.GetMinioPath(jobName, CodeMountPath + "/"),
+					MountPath: CodeMountPath,
+					ReadOnly:  false,
+				},
+			},
+			{
+				HostPath: models.StHostPath{
+					Path:      dataActualPath,
+					MountPath: DataSetMountPath,
+					ReadOnly:  true,
+				},
+			},
+			{
+				HostPath: models.StHostPath{
+					Path:      storage.GetMinioPath(jobName, ModelMountPath + "/"),
+					MountPath: ModelMountPath,
+					ReadOnly:  false,
+				},
+			},
+			{
+				HostPath: models.StHostPath{
+					Path:      storage.GetMinioPath(jobName, BenchMarkMountPath + "/"),
+					MountPath: BenchMarkMountPath,
+					ReadOnly:  true,
+				},
+			},
+			{
+				HostPath: models.StHostPath{
+					Path:      storage.GetMinioPath(jobName, Snn4imagenetMountPath + "/"),
+					MountPath: Snn4imagenetMountPath,
+					ReadOnly:  true,
+				},
+			},
+			{
+				HostPath: models.StHostPath{
+					Path:      storage.GetMinioPath(jobName, BrainScoreMountPath + "/"),
+					MountPath: BrainScoreMountPath,
+					ReadOnly:  true,
+				},
+			},
+		},
 	})
+	if err != nil {
+		log.Error("CreateJob failed:", err.Error(), ctx.Data["MsgID"])
+		return err
+	}
+	if jobResult.Code != Success {
+		log.Error("CreateJob(%s) failed:%s", jobName, jobResult.Msg, ctx.Data["MsgID"])
+		return errors.New(jobResult.Msg)
+	}
+
+	var jobID = jobResult.Payload["jobId"].(string)
+	task.JobID = jobID
+	task.Status = string(models.JobWaiting)
+	err =  models.UpdateJob(task)
 
 	if err != nil {
+		log.Error("UpdateJob(%s) failed:%v", jobName, err.Error(), ctx.Data["MsgID"])
 		return err
 	}
 

modules/modelarts/modelarts.go

@@ -277,7 +277,7 @@ func GenerateTrainJob(ctx *context.Context, req *GenerateTrainJobReq) (err error
 		DatasetName:       attach.Name,
 		CommitID:          req.CommitID,
 		IsLatestVersion:   req.IsLatestVersion,
-		ComputeResource:   NPUResource,
+		//ComputeResource:   NPUResource,
 		EngineID:          req.EngineID,
 		TrainUrl:          req.TrainUrl,
 		BranchName:        req.BranchName,

modules/modelarts/resty.go

@@ -174,7 +174,7 @@ sendjob:
 	return &result, nil
 }
 
-func StopJob(jobID string, param models.NotebookAction) (*models.NotebookActionResult, error) {
+func ManageNotebook(jobID string, param models.NotebookAction) (*models.NotebookActionResult, error) {
 	checkSetting()
 	client := getRestyClient()
 	var result models.NotebookActionResult
@@ -207,8 +207,8 @@ sendjob:
 	}
 
 	if len(response.ErrorCode) != 0 {
-		log.Error("StopJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
-		return &result, fmt.Errorf("StopJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
+		log.Error("ManageNotebook failed(%s): %s", response.ErrorCode, response.ErrorMsg)
+		return &result, fmt.Errorf("ManageNotebook failed(%s): %s", response.ErrorCode, response.ErrorMsg)
 	}
 
 	return &result, nil

modules/storage/minio.go

@@ -11,6 +11,7 @@ import (
 	"strings"
 	"time"
 
+	"code.gitea.io/gitea/modules/setting"
 	"github.com/minio/minio-go"
 )
 
@@ -128,3 +129,7 @@ func (m *MinioStorage) UploadObject(fileName, filePath string) error {
 	_, err := m.client.FPutObject(m.bucket, fileName, filePath, minio.PutObjectOptions{})
 	return err
 }
+
+func GetMinioPath(jobName, suffixPath string) string {
+	return setting.Attachment.Minio.RealPath + setting.Attachment.Minio.Bucket + "/" + setting.CBCodePathPrefix + jobName + suffixPath
+}

options/locale/locale_zh-CN.ini

@@ -787,6 +787,7 @@ model_noright=无权限操作
 model_rename=模型名称重复，请修改模型名称
 
 debug=调试
+debug_again=再次调试
 stop=停止
 delete=删除
 model_download=模型下载

routers/repo/cloudbrain.go

@@ -206,7 +206,7 @@ func CloudBrainCreate(ctx *context.Context, form auth.CreateCloudBrainForm) {
 	}
 	repo := ctx.Repo.Repository
 	downloadCode(repo, codePath)
-	uploadCodeToMinio(codePath+"/", jobName, "/code/")
+	uploadCodeToMinio(codePath+"/", jobName, cloudbrain.CodeMountPath+"/")
 
 	modelPath := setting.JobPath + jobName + cloudbrain.ModelMountPath + "/"
 	mkModelPath(modelPath)
@@ -236,9 +236,10 @@ func CloudBrainCreate(ctx *context.Context, form auth.CreateCloudBrainForm) {
 		uploadCodeToMinio(brainScorePath+"/", jobName, cloudbrain.BrainScoreMountPath+"/")
 	}
 
-	err = cloudbrain.GenerateTask(ctx, jobName, image, command, uuid, codePath, getMinioPath(jobName, cloudbrain.ModelMountPath+"/"),
-		getMinioPath(jobName, cloudbrain.BenchMarkMountPath+"/"), getMinioPath(jobName, cloudbrain.Snn4imagenetMountPath+"/"),
-		getMinioPath(jobName, cloudbrain.BrainScoreMountPath+"/"), jobType, gpuQueue, resourceSpecId)
+	err = cloudbrain.GenerateTask(ctx, jobName, image, command, uuid, storage.GetMinioPath(jobName, cloudbrain.CodeMountPath+"/"),
+		storage.GetMinioPath(jobName, cloudbrain.ModelMountPath+"/"),
+		storage.GetMinioPath(jobName, cloudbrain.BenchMarkMountPath+"/"), storage.GetMinioPath(jobName, cloudbrain.Snn4imagenetMountPath+"/"),
+		storage.GetMinioPath(jobName, cloudbrain.BrainScoreMountPath+"/"), jobType, gpuQueue, resourceSpecId)
 	if err != nil {
 		cloudBrainNewDataPrepare(ctx)
 		ctx.RenderWithErr(err.Error(), tplCloudBrainNew, &form)
@@ -247,6 +248,72 @@ func CloudBrainCreate(ctx *context.Context, form auth.CreateCloudBrainForm) {
 	ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob")
 }
 
+func CloudBrainRestart(ctx *context.Context) {
+	var jobID = ctx.Params(":jobid")
+	var resultCode = "0"
+	var errorMsg = ""
+	var status = ""
+
+	for {
+		task, err := models.GetCloudbrainByJobID(jobID)
+		if err != nil {
+			log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error(), ctx.Data["MsgID"])
+			resultCode = "-1"
+			errorMsg = "system error"
+			break
+		}
+
+		if task.Status != string(models.JobStopped) && task.Status != string(models.JobSucceeded) && task.Status != string(models.JobFailed) {
+			log.Error("the job(%s) is not stopped", task.JobName, ctx.Data["MsgID"])
+			resultCode = "-1"
+			errorMsg = "the job is not stopped"
+			break
+		}
+
+		if task.Image == "" || task.GpuQueue == "" || task.Type != models.TypeCloudBrainOne {
+			log.Error("the job(%s) version is too old", task.JobName, ctx.Data["MsgID"])
+			resultCode = "-1"
+			errorMsg = "the job's version is too old and can not be restarted"
+			break
+		}
+
+		count, err := models.GetCloudbrainCountByUserID(ctx.User.ID)
+		if err != nil {
+			log.Error("GetCloudbrainCountByUserID failed:%v", err, ctx.Data["MsgID"])
+			resultCode = "-1"
+			errorMsg = "system error"
+			break
+		} else {
+			if count >= 1 {
+				log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
+				resultCode = "-1"
+				errorMsg = "the user already has running or waiting task"
+				break
+			}
+		}
+
+		err = cloudbrain.RestartTask(ctx, task)
+		if err != nil {
+			log.Error("RestartTask failed:%v", err.Error(), ctx.Data["MsgID"])
+			resultCode = "-1"
+			errorMsg = "system error"
+			break
+		}
+
+		status = task.Status
+		jobID = task.JobID
+
+		break
+	}
+
+	ctx.JSON(200, map[string]string{
+		"result_code": resultCode,
+		"error_msg":   errorMsg,
+		"status":      status,
+		"job_id":      jobID,
+	})
+}
+
 func CloudBrainShow(ctx *context.Context) {
 	ctx.Data["PageIsCloudBrain"] = true
 
@@ -351,32 +418,53 @@ func CloudBrainCommitImage(ctx *context.Context, form auth.CommitImageCloudBrain
 
 func CloudBrainStop(ctx *context.Context) {
 	var jobID = ctx.Params(":jobid")
-	task, err := models.GetCloudbrainByJobID(jobID)
-	if err != nil {
-		ctx.ServerError("GetCloudbrainByJobID failed", err)
-		return
-	}
+	var resultCode = "0"
+	var errorMsg = ""
+	var status = ""
 
-	if task.Status == string(models.JobStopped) || task.Status == string(models.JobFailed) {
-		log.Error("the job(%s) has been stopped", task.JobName, ctx.Data["msgID"])
-		ctx.ServerError("the job has been stopped", errors.New("the job has been stopped"))
-		return
-	}
+	for {
+		task, err := models.GetCloudbrainByJobID(jobID)
+		if err != nil {
+			log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err, ctx.Data["msgID"])
+			resultCode = "-1"
+			errorMsg = "system error"
+			break
+		}
 
-	err = cloudbrain.StopJob(jobID)
-	if err != nil {
-		log.Error("StopJob(%s) failed:%v", task.JobName, err.Error(), ctx.Data["msgID"])
-		ctx.ServerError("StopJob failed", err)
-		return
-	}
+		if task.Status == string(models.JobStopped) || task.Status == string(models.JobFailed) {
+			log.Error("the job(%s) has been stopped", task.JobName, ctx.Data["msgID"])
+			resultCode = "-1"
+			errorMsg = "system error"
+			break
+		}
 
-	task.Status = string(models.JobStopped)
-	err = models.UpdateJob(task)
-	if err != nil {
-		ctx.ServerError("UpdateJob failed", err)
-		return
+		err = cloudbrain.StopJob(jobID)
+		if err != nil {
+			log.Error("StopJob(%s) failed:%v", task.JobName, err, ctx.Data["msgID"])
+			resultCode = "-1"
+			errorMsg = "system error"
+			break
+		}
+
+		task.Status = string(models.JobStopped)
+		err = models.UpdateJob(task)
+		if err != nil {
+			log.Error("UpdateJob(%s) failed:%v", task.JobName, err, ctx.Data["msgID"])
+			resultCode = "-1"
+			errorMsg = "system error"
+			break
+		}
+
+		status = task.Status
+		break
 	}
-	ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob")
+
+	ctx.JSON(200, map[string]string{
+		"result_code": resultCode,
+		"error_msg":   errorMsg,
+		"status":      status,
+		"job_id":      jobID,
+	})
 }
 
 func StopJobsByUserID(userID int64) {
@@ -423,7 +511,7 @@ func StopJobs(cloudBrains []*models.Cloudbrain) {
 					Action: models.ActionStop,
 				}
 				err := retry(3, time.Second*30, func() error {
-					_, err := modelarts.StopJob(taskInfo.JobID, param)
+					_, err := modelarts.ManageNotebook(taskInfo.JobID, param)
 					return err
 				})
 				logErrorAndUpdateJobStatus(err, taskInfo)
@@ -560,7 +648,7 @@ func getImages(ctx *context.Context, imageType string) {
 
 func GetModelDirs(jobName string, parentDir string) (string, error) {
 	var req string
-	modelActualPath := getMinioPath(jobName, cloudbrain.ModelMountPath+"/")
+	modelActualPath := storage.GetMinioPath(jobName, cloudbrain.ModelMountPath+"/")
 	if parentDir == "" {
 		req = "baseDir=" + modelActualPath
 	} else {
@@ -570,10 +658,6 @@ func GetModelDirs(jobName string, parentDir string) (string, error) {
 	return getDirs(req)
 }
 
-func getMinioPath(jobName, suffixPath string) string {
-	return setting.Attachment.Minio.RealPath + setting.Attachment.Minio.Bucket + "/" + setting.CBCodePathPrefix + jobName + suffixPath
-}
-
 func CloudBrainDownloadModel(ctx *context.Context) {
 	parentDir := ctx.Query("parentDir")
 	fileName := ctx.Query("fileName")

routers/repo/modelarts.go

@@ -42,6 +42,7 @@ const (
 
 func DebugJobIndex(ctx *context.Context) {
 	debugListType := ctx.Query("debugListType")
+	ctx.Data["ListType"] = debugListType
 	MustEnableCloudbrain(ctx)
 	repo := ctx.Repo.Repository
 	page := ctx.QueryInt("page")
@@ -73,21 +74,19 @@ func DebugJobIndex(ctx *context.Context) {
 	}
 
 	for i, task := range ciTasks {
+		ciTasks[i].CanDebug = cloudbrain.CanCreateOrDebugJob(ctx)
+		ciTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
+
 		if task.Cloudbrain.Type == models.TypeCloudBrainOne {
-			ciTasks[i].CanDebug = cloudbrain.CanCreateOrDebugJob(ctx)
-			ciTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
 			ciTasks[i].Cloudbrain.ComputeResource = modelarts.GPUResource
-		}
-		if task.Cloudbrain.Type == models.TypeCloudBrainTwo {
-			ciTasks[i].CanDebug = cloudbrain.CanCreateOrDebugJob(ctx)
-			ciTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
+		} else if task.Cloudbrain.Type == models.TypeCloudBrainTwo {
 			ciTasks[i].Cloudbrain.ComputeResource = modelarts.NPUResource
 		}
-
 	}
 
 	pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)
-	pager.SetDefaultParams(ctx)
+	//pager.SetDefaultParams(ctx)
+	pager.AddParam(ctx, "debugListType", "ListType")
 	ctx.Data["Page"] = pager
 	ctx.Data["PageIsCloudBrain"] = true
 	ctx.Data["Tasks"] = ciTasks
@@ -232,38 +231,91 @@ func NotebookDebug(ctx *context.Context) {
 	ctx.Redirect(debugUrl)
 }
 
-func NotebookStop(ctx *context.Context) {
+func NotebookManage(ctx *context.Context) {
 	var jobID = ctx.Params(":jobid")
-	log.Info(jobID)
-	task, err := models.GetCloudbrainByJobID(jobID)
-	if err != nil {
-		ctx.ServerError("GetCloudbrainByJobID failed", err)
-		return
-	}
+	var action = ctx.Params(":action")
+	var resultCode = "0"
+	var errorMsg = ""
+	var status = ""
 
-	if task.Status != string(models.JobRunning) {
-		log.Error("the job(%s) is not running", task.JobName)
-		ctx.ServerError("the job is not running", errors.New("the job is not running"))
-		return
-	}
+	for {
+		task, err := models.GetCloudbrainByJobID(jobID)
+		if err != nil {
+			log.Error("GetCloudbrainByJobID failed:%v", err, ctx.Data["MsgID"])
+			resultCode = "-1"
+			errorMsg = "system error"
+			break
+		}
 
-	param := models.NotebookAction{
-		Action: models.ActionStop,
-	}
-	res, err := modelarts.StopJob(jobID, param)
-	if err != nil {
-		log.Error("StopJob(%s) failed:%v", task.JobName, err.Error())
-		ctx.ServerError("StopJob failed", err)
-		return
-	}
+		if action == models.ActionStop {
+			if task.Status != string(models.ModelArtsRunning) {
+				log.Error("the job(%s) is not running", task.JobName, ctx.Data["MsgID"])
+				resultCode = "-1"
+				errorMsg = "the job is not running"
+				break
+			}
+		} else if action == models.ActionRestart {
+			if task.Status != string(models.ModelArtsStopped) && task.Status != string(models.ModelArtsStartFailed) && task.Status != string(models.ModelArtsCreateFailed) {
+				log.Error("the job(%s) is not stopped", task.JobName, ctx.Data["MsgID"])
+				resultCode = "-1"
+				errorMsg = "the job is not stopped"
+				break
+			}
 
-	task.Status = res.CurrentStatus
-	err = models.UpdateJob(task)
-	if err != nil {
-		ctx.ServerError("UpdateJob failed", err)
-		return
+			count, err := models.GetCloudbrainNotebookCountByUserID(ctx.User.ID)
+			if err != nil {
+				log.Error("GetCloudbrainNotebookCountByUserID failed:%v", err, ctx.Data["MsgID"])
+				resultCode = "-1"
+				errorMsg = "system error"
+				break
+			} else {
+				if count >= 1 {
+					log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
+					resultCode = "-1"
+					errorMsg = "you have already a running or waiting task, can not create more"
+					break
+				}
+			}
+
+			action = models.ActionStart
+		} else {
+			log.Error("the action(%s) is illegal", action, ctx.Data["MsgID"])
+			resultCode = "-1"
+			errorMsg = "非法操作"
+			break
+		}
+
+		param := models.NotebookAction{
+			Action: action,
+		}
+		res, err := modelarts.ManageNotebook(jobID, param)
+		if err != nil {
+			log.Error("ManageNotebook(%s) failed:%v", task.JobName, err.Error(), ctx.Data["MsgID"])
+			resultCode = "-1"
+			errorMsg = "启动失败"
+			break
+		}
+
+		task.Status = res.CurrentStatus
+		err = models.UpdateJob(task)
+		if err != nil {
+			log.Error("UpdateJob(%s) failed:%v", task.JobName, err.Error(), ctx.Data["MsgID"])
+			resultCode = "-1"
+			errorMsg = "system error"
+			break
+		}
+
+		status = task.Status
+
+		break
 	}
-	ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob")
+
+	ctx.JSON(200, map[string]string{
+		"result_code": resultCode,
+		"error_msg":   errorMsg,
+		"status":      status,
+		"job_id":      jobID,
+	})
 }
 
 func NotebookDel(ctx *context.Context) {
@@ -323,6 +375,7 @@ func TrainJobIndex(ctx *context.Context) {
 	for i, task := range tasks {
 		tasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
 		tasks[i].CanModify = cloudbrain.CanModifyJob(ctx, &task.Cloudbrain)
+		tasks[i].ComputeResource = modelarts.NPUResource
 	}
 
 	pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)

routers/routes/routes.go

@@ -968,6 +968,7 @@ func RegisterRoutes(m *macaron.Macaron) {
 				m.Post("/commit_image", cloudbrain.AdminOrOwnerOrJobCreaterRight, bindIgnErr(auth.CommitImageCloudBrainForm{}), repo.CloudBrainCommitImage)
 				m.Post("/stop", cloudbrain.AdminOrOwnerOrJobCreaterRight, repo.CloudBrainStop)
 				m.Post("/del", cloudbrain.AdminOrOwnerOrJobCreaterRight, repo.CloudBrainDel)
+				m.Post("/restart", cloudbrain.AdminOrOwnerOrJobCreaterRight, repo.CloudBrainRestart)
 				m.Get("/rate", reqRepoCloudBrainReader, repo.GetRate)
 				m.Get("/models", reqRepoCloudBrainReader, repo.CloudBrainShowModels)
 				m.Get("/download_model", cloudbrain.AdminOrOwnerOrJobCreaterRight, repo.CloudBrainDownloadModel)
@@ -1003,7 +1004,7 @@ func RegisterRoutes(m *macaron.Macaron) {
 				m.Group("/:jobid", func() {
 					m.Get("", reqRepoCloudBrainReader, repo.NotebookShow)
 					m.Get("/debug", reqRepoCloudBrainWriter, repo.NotebookDebug)
-					m.Post("/stop", cloudbrain.AdminOrOwnerOrJobCreaterRight, repo.NotebookStop)
+					m.Post("/:action", cloudbrain.AdminOrOwnerOrJobCreaterRight, repo.NotebookManage)
 					m.Post("/del", cloudbrain.AdminOrOwnerOrJobCreaterRight, repo.NotebookDel)
 				})
 				m.Get("/create", reqRepoCloudBrainWriter, repo.NotebookNew)

templates/repo/debugjob/index.tmpl

@@ -202,14 +202,20 @@
         <div class="rect5"></div>
     </div>
 </div>
-<!-- 提示框 -->
-<div class="alert"></div>
+
 
 <div class="repository release dataset-list view">
     {{template "repo/header" .}}
+    <!-- {{template "base/alert" .}} -->
+    <!-- 提示框 -->
+    
     <!-- 列表容器 -->
     <div class="ui container">
-        <div class="ui two column stackable grid ">
+        <div class="ui negative message" style="display: none;">
+            <i class="close icon"></i>
+            <p></p>
+        </div>
+        <div class="ui two column stackable grid">
                 <div class="column">
                     <div class="ui blue small menu compact selectcloudbrain">
                         <a class="active item" href="{{.RepoLink}}/debugjob?debugListType=all">{{$.i18n.Tr "repo.modelarts.notebook"}}</a>
@@ -282,7 +288,7 @@
                             <div class="row">
                                 <!-- 任务名 -->
                                 <div class="four wide column">
-                                    <a class="title" href='{{if eq .ComputeResource "CPU/GPU"}}{{$.RepoLink}}/cloudbrain{{else}}{{$.RepoLink}}/modelarts/notebook{{end}}/{{.JobID}}' title="{{.JobName}}" style="font-size: 14px;">
+                                    <a class="title" href='{{if eq .ComputeResource "CPU/GPU"}}{{$.RepoLink}}/cloudbrain{{else}}{{$.RepoLink}}/modelarts/notebook{{end}}/{{.JobID}}'  title="{{.JobName}}" style="font-size: 14px;">
                                         <span class="fitted text_over" style="width: 90%;vertical-align: middle;">{{.JobName}}</span>
                                     </a>
                                 </div>
@@ -315,34 +321,44 @@
                                         </a>
                                         {{end}} -->
                                         <!-- 调试 -->
-                                        {{if .CanDebug}}
-                                            {{if eq .ComputeResource "CPU/GPU"}}
-                                            <a id="model-debug-{{.JobID}}" class='ui basic {{if ne .Status "RUNNING"}} disabled {{else}}blue {{end}}button' href="{{$.RepoLink}}/cloudbrain/{{.JobID}}/debug" target="_blank">
-                                                {{$.i18n.Tr "repo.debug"}}
-                                            </a>
+                                        <form id="debugAgainForm-{{.JobID}}">
+                                            {{$.CsrfTokenHtml}}
+                                            {{if .CanDebug}}
+                                                {{if eq .Status "RUNNING"}}
+                                                <a style="margin: 0 1rem;" id="model-debug-{{.JobID}}" class='ui basic blue button' onclick='debugAgain("{{.JobID}}","{{if eq .ComputeResource "CPU/GPU"}}{{$.RepoLink}}/cloudbrain{{else}}{{$.RepoLink}}/modelarts/notebook{{end}}/{{.JobID}}/")'>
+                                                    {{$.i18n.Tr "repo.debug"}}
+                                                </a>
+                                                {{else}}
+                                                <a id="model-debug-{{.JobID}}" class='ui basic {{if eq .Status "CREATING" "STOPPING"  "WAITING" "STARTING"}} disabled {{else}}blue {{end}}button' onclick='debugAgain("{{.JobID}}","{{if eq .ComputeResource "CPU/GPU"}}{{$.RepoLink}}/cloudbrain{{else}}{{$.RepoLink}}/modelarts/notebook{{end}}/{{.JobID}}/")'>
+                                                    {{$.i18n.Tr "repo.debug_again"}}
+                                                </a>
+                                                {{end}}
                                             {{else}}
-                                            <a id="model-debug-{{.JobID}}" class='ui basic {{if ne .Status "RUNNING"}} disabled {{else}}blue {{end}}button' href="{{$.RepoLink}}/modelarts/notebook/{{.JobID}}/debug" target="_blank">
-                                                {{$.i18n.Tr "repo.debug"}}
+                                            
+                                            <a  class="ui basic disabled button">
+                                                {{$.i18n.Tr "repo.debug_again"}}
                                             </a>
                                             {{end}}
-                                        {{else}}
-                                        <a  class="ui basic disabled button">
-                                            {{$.i18n.Tr "repo.debug"}}
-                                        </a>
-                                        {{end}}
+                                        </form>
+                                        
                                         <!-- 停止 -->
-                                        <form id="stopForm-{{.JobID}}" action="{{if eq .ComputeResource "CPU/GPU"}}{{$.RepoLink}}/cloudbrain{{else}}{{$.RepoLink}}/modelarts/notebook{{end}}/{{.JobID}}/stop" method="post" style="margin-left:-1px;">
+                                        <form id="stopForm-{{.JobID}}"  style="margin-left:-1px;">
                                             {{$.CsrfTokenHtml}}
                                             {{if .CanDel}}
-                                            <a id="stop-model-debug-{{.JobID}}" class='ui basic {{if eq .Status "STOPPED" "FAILED" "START_FAILED" "STOPPING"  "CREATING"  "STARTING"}}disabled {{else}}blue {{end}}button' onclick="document.getElementById('stopForm-{{.JobID}}').submit();">
-                                                {{$.i18n.Tr "repo.stop"}}
-                                            </a>
+                                                {{if eq .ComputeResource "CPU/GPU" }}
+                                                <a id="stop-model-debug-{{.JobID}}" class='ui basic {{if eq .Status "STOPPED" "FAILED" "START_FAILED" "STOPPING"  "CREATING"  "STARTING"}}disabled {{else}}blue {{end}}button' onclick='stopDebug("{{.JobID}}","{{$.RepoLink}}/cloudbrain/{{.JobID}}/stop")'>
+                                                    {{$.i18n.Tr "repo.stop"}}
+                                                </a>
+                                                {{else}}
+                                                <a id="stop-model-debug-{{.JobID}}" class='ui basic {{if eq .Status "STOPPED" "FAILED" "START_FAILED" "STOPPING"  "CREATING"  "STARTING"}}disabled {{else}}blue {{end}}button' onclick='stopDebug("{{.JobID}}","{{$.RepoLink}}/modelarts/notebook/{{.JobID}}/stop")'>
+                                                    {{$.i18n.Tr "repo.stop"}}
+                                                </a>
+                                                {{end}}
                                             {{else}}
-                                            <a  class="ui basic disabled button" onclick="document.getElementById('stopForm-{{.JobID}}').submit();">
+                                            <a  class="ui basic disabled button">
                                                     {{$.i18n.Tr "repo.stop"}}
                                             </a>
                                             {{end}}
-                                            <input type="hidden" name="debugListType" value="all">
                                         </form>
                                         <!-- 删除 -->
                                         <form  id="delForm-{{.JobID}}" action="{{if eq .ComputeResource "CPU/GPU"}}{{$.RepoLink}}/cloudbrain{{else}}{{$.RepoLink}}/modelarts/notebook{{end}}/{{.JobID}}/del" method="post">
@@ -463,10 +479,21 @@
 
 <script>
     // 调试和评分新开窗口
+    const {AppSubUrl, StaticUrlPrefix, csrf} = window.config;
     let url={{.RepoLink}}
-    let getParam=location.search.split('?debugListType=').pop()
+    let getParam=getQueryVariable('debugListType')
     let dropdownValue = getParam==='all'||getParam==='' ? '全部' : getParam
     localStorage.setItem('all',location.href)
+    function getQueryVariable(variable)
+    {
+        let query = window.location.search.substring(1);
+        let vars = query.split("&");
+        for (let i=0;i<vars.length;i++) {
+                let pair = vars[i].split("=");
+                if(pair[0] == variable){return pair[1];}
+        }
+        return(false);
+    }
     function stop(obj) {
         if (obj.style.color != "rgb(204, 204, 204)") {
             obj.target = '_blank'
@@ -499,7 +526,68 @@
                 .modal('show')
         }
     }
+    function debugAgain(JobID,debugUrl){
+        if($('#' + JobID+ '-text').text()==="RUNNING"){
+            window.open(debugUrl+'debug')
+        }else{
+            $.ajax({
+                type:"POST",
+                url:debugUrl+'restart',
+                data:$('#debugAgainForm-'+JobID).serialize(),
+                success:function(res){
+                    if(res.result_code==="0"){
+                        if(res.job_id!==JobID){
+                            location.reload()
+                        }else{
+                            $('#' + JobID+'-icon').removeClass().addClass(res.status)
+                            $('#' + JobID+ '-text').text(res.status)
+                            $('#model-debug-'+JobID).removeClass('blue').addClass('disabled')
+                            $('#model-delete-'+JobID).removeClass('blue').addClass('disabled')
+                        }
+                    }else{
+                        $(".ui.negative.message").css("display","block")
+                        $(".ui.negative.message p").text(res.error_msg)
+                        setTimeout("$('.message .close').click()",3000)
+                    }
+                },
+                error :function(res){
+                    console.log(res)
 
+                }
+            })
+        }
+        
+    }
+    function stopDebug(JobID,stopUrl){
+        $.ajax({
+            type:"POST",
+            url:stopUrl,
+            data:$('#stopForm-'+JobID).serialize(),
+            success:function(res){
+                if(res.result_code==="0"){
+                    $('#' + JobID+'-icon').removeClass().addClass(res.status)
+                    $('#' + JobID+ '-text').text(res.status)
+                    if(res.status==="STOPPED"){
+                        $('#model-debug-'+JobID).removeClass('blue').addClass('disabled').text("再次调试").css("margin","0")
+                        $('#model-image-'+JobID).removeClass('blue').addClass('disabled')
+                        $('#stop-model-debug-'+JobID).removeClass('blue').addClass('disabled')
+                    }
+                    else{
+                        $('#model-debug-'+JobID).removeClass('blue').addClass('disabled')
+                        $('#stop-model-debug-'+JobID).removeClass('blue').addClass('disabled')
+                    }
+                    
+                }else{
+                    $("ui.negative.message").text(res.error_msg)
+                }
+            },
+            error :function(res){
+                console.log(res)
+
+            }
+        })
+
+    }
     // 加载任务状态
     var timeid = window.setInterval(loadJobStatus, 15000);
     $(document).ready(loadJobStatus);
@@ -508,8 +596,9 @@
             const jobID = job.dataset.jobid;
             const repoPath = job.dataset.repopath;
             const computeResource = job.dataset.resource
-            const initArray = ['STOPPED','FAILED','START_FAILED','CREATE_FAILED']
+            const initArray = ['STOPPED','FAILED','START_FAILED','CREATE_FAILED','SUCCEEDED']
             if (initArray.includes(job.textContent.trim())) {
+                
                 return
             }
             const diffResource = computeResource == "NPU" ? 'modelarts/notebook' : 'cloudbrain'
@@ -521,32 +610,30 @@
                     $('#' + jobID+ '-text').text(status)
                 }
                 if(status==="RUNNING"){
-                    $('#model-debug-'+jobID).removeClass('disabled')
-                    $('#model-debug-'+jobID).addClass('blue')
-                    $('#model-image-'+jobID).removeClass('disabled')
-                    $('#model-image-'+jobID).addClass('blue')
+                    $('#model-debug-'+jobID).removeClass('disabled').addClass('blue').text('调试').css("margin","0 1rem")
+                    $('#model-image-'+jobID).removeClass('disabled').addClass('blue')
                 }
                 if(status!=="RUNNING"){
-                    $('#model-debug-'+jobID).removeClass('blue')
-                    $('#model-debug-'+jobID).addClass('disabled')
-                    $('#model-image-'+jobID).removeClass('blue')
-                    $('#model-image-'+jobID).addClass('disabled')
-
+                    // $('#model-debug-'+jobID).removeClass('blue')
+                    // $('#model-debug-'+jobID).addClass('disabled')
+                    $('#model-image-'+jobID).removeClass('blue').addClass('disabled')
+                }
+                if(["CREATING","STOPPING","WAITING","STARTING"].includes(status)){
+                    $('#model-debug-'+jobID).removeClass('blue').addClass('disabled')
+                }
+                if(['STOPPED','FAILED','START_FAILED','CREATE_FAILED','SUCCEEDED'].includes(status)){
+                    $('#model-debug-'+jobID).removeClass('disabled').addClass('blue').text('再次调试').css("margin","0") 
                 }
                 if(["RUNNING","WAITING"].includes(status)){
-                   $('#stop-model-debug-'+jobID).removeClass('disabled')
-                   $('#stop-model-debug-'+jobID).addClass('blue')
+                   $('#stop-model-debug-'+jobID).removeClass('disabled').addClass('blue')
                 }
                 if(["CREATING","STOPPING","STARTING","STOPPED","FAILED","START_FAILED"].includes(status)){
-                   $('#stop-model-debug-'+jobID).removeClass('blue')
-                   $('#stop-model-debug-'+jobID).addClass('disabled')
+                   $('#stop-model-debug-'+jobID).removeClass('blue').addClass('disabled')
                 }
                 if(status==="STOPPED" || status==="FAILED"|| status==="START_FAILED"){
-                    $('#model-delete-'+jobID).removeClass('disabled')
-                    $('#model-delete-'+jobID).addClass('blue')
+                    $('#model-delete-'+jobID).removeClass('disabled').addClass('blue')
                 }else{
-                    $('#model-delete-'+jobID).removeClass('blue')
-                    $('#model-delete-'+jobID).addClass('disabled')
+                    $('#model-delete-'+jobID).removeClass('blue').addClass('disabled')
                 }
             }).fail(function(err) {
                 console.log(err);
@@ -554,6 +641,7 @@
         });
     };
     $(document).ready(function(){
+        dropdownValue = dropdownValue==="CPU%2FGPU"? 'CPU/GPU' : dropdownValue
         $('.default.text').text(dropdownValue)
         $('.ui.dropdown')
         .dropdown({
@@ -564,6 +652,12 @@
                 location.href = `${url}/debugjob?debugListType=${value}`
             }
         })
+        $('.message .close')
+        .on('click', function() {
+            $(this)
+            .closest('.message')
+            .transition('fade')
+        })
     })
     
     
@@ -601,7 +695,6 @@
     // 显示弹窗，弹出相应的信息
     function showmask() {
         var image_tag = !$('#image_tag').val()
-        console.log("image_tag",image_tag)
         if(image_tag){
             return
         }

templates/repo/header.tmpl

@@ -149,6 +149,12 @@
 					{{.i18n.Tr "repo.model_manager"}}
 				</a>
 				{{end}}
+				{{if .Permission.CanRead $.UnitTypeModelManage}}
+				<a class="{{if .isModelManage}}active{{end}} item" href="{{.RepoLink}}/modelmanage/show_model">
+					<svg class="svg" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" width="16" height="16"><path fill="none" d="M0 0h24v24H0z"/><path d="M3.741 1.408l18.462 10.154a.5.5 0 0 1 0 .876L3.741 22.592A.5.5 0 0 1 3 22.154V1.846a.5.5 0 0 1 .741-.438zM5 13v6.617L18.85 12 5 4.383V11h5v2H5z"/></svg>
+					{{.i18n.Tr "repo.model_manager"}}
+				</a>
+				{{end}}
 				{{if .Permission.CanRead $.UnitTypeCloudBrain}}
 				<a class="{{if .PageIsCloudBrain}}active{{end}} item" href="{{.RepoLink}}/debugjob?debugListType=all">
 					<span>