Merge pull request 'check bootfile' (#2672) from fix-2559 into V20220815

Reviewed-on: https://git.openi.org.cn/OpenI/aiforge/pulls/2672
Reviewed-by: lewis <747342561@qq.com>

options/locale/locale_en-US.ini

@@ -1073,6 +1073,7 @@ cloudbrain_operate = Operate
 cloudbrain_status_createtime = Status/Createtime
 cloudbrain_status_runtime = Running Time
 cloudbrain_jobname_err=Name must start with a lowercase letter or number,can include lowercase letter,number,_ and -,can not end with _,  and can be up to 36 characters long.
+cloudbrain_bootfile_err=The bootfile does not exist in the repository
 cloudbrain_query_fail=Failed to query cloudbrain information.
 cloudbrain.mirror_tag = Mirror Tag
 cloudbrain.mirror_description = Mirror Description

options/locale/locale_zh-CN.ini

@@ -1076,6 +1076,7 @@ cloudbrain_operate=操作
 cloudbrain_status_createtime=状态/创建时间
 cloudbrain_status_runtime = 运行时长
 cloudbrain_jobname_err=只能以小写字母或数字开头且只包含小写字母、数字、_和-，不能以_结尾，最长36个字符。
+cloudbrain_bootfile_err=仓库中不存在启动文件
 cloudbrain_query_fail=查询云脑任务失败。
 cloudbrain.mirror_tag = 镜像标签
 cloudbrain.mirror_description = 镜像描述

routers/repo/cloudbrain.go

@@ -239,6 +239,7 @@ func CloudBrainCreate(ctx *context.Context, form auth.CreateCloudBrainForm) {
 	codePath := setting.JobPath + jobName + cloudbrain.CodeMountPath
 	resourceSpecId := form.ResourceSpecId
 	branchName := form.BranchName
+	bootFile := strings.TrimSpace(form.BootFile)
 	repo := ctx.Repo.Repository
 	tpl := tplCloudBrainNew
 
@@ -305,6 +306,13 @@ func CloudBrainCreate(ctx *context.Context, form auth.CreateCloudBrainForm) {
 
 	command := cloudbrain.GetCloudbrainDebugCommand()
 	if jobType == string(models.JobTypeTrain) {
+		bootFileExist, err := ctx.Repo.FileExists(bootFile, branchName)
+		if err != nil || !bootFileExist {
+			log.Error("Get bootfile error:", err, ctx.Data["MsgID"])
+			cloudBrainNewDataPrepare(ctx)
+			ctx.RenderWithErr(ctx.Tr("repo.cloudbrain_bootfile_err"), tpl, &form)
+			return
+		}
 		tpl = tplCloudBrainTrainJobNew
 		commandTrain, err := getTrainJobCommand(form)
 		if err != nil {
@@ -413,6 +421,7 @@ func CloudBrainInferenceJobCreate(ctx *context.Context, form auth.CreateCloudBra
 	codePath := setting.JobPath + jobName + cloudbrain.CodeMountPath
 	resourceSpecId := form.ResourceSpecId
 	branchName := form.BranchName
+	bootFile := strings.TrimSpace(form.BootFile)
 	labelName := form.LabelName
 	repo := ctx.Repo.Repository
 
@@ -450,6 +459,14 @@ func CloudBrainInferenceJobCreate(ctx *context.Context, form auth.CreateCloudBra
 		return
 	}
 
+	bootFileExist, err := ctx.Repo.FileExists(bootFile, branchName)
+	if err != nil || !bootFileExist {
+		log.Error("Get bootfile error:", err, ctx.Data["MsgID"])
+		cloudBrainNewDataPrepare(ctx)
+		ctx.RenderWithErr(ctx.Tr("repo.cloudbrain_bootfile_err"), tpl, &form)
+		return
+	}
+
 	count, err := models.GetCloudbrainCountByUserID(ctx.User.ID, jobType)
 	if err != nil {
 		log.Error("GetCloudbrainCountByUserID failed:%v", err, ctx.Data["MsgID"])

routers/repo/grampus.go

@@ -215,6 +215,14 @@ func GrampusTrainJobGpuCreate(ctx *context.Context, form auth.CreateGrampusTrain
 		return
 	}
 
+	bootFileExist, err := ctx.Repo.FileExists(bootFile, branchName)
+	if err != nil || !bootFileExist {
+		log.Error("Get bootfile error:", err, ctx.Data["MsgID"])
+		grampusTrainJobNewDataPrepare(ctx, grampus.ProcessorTypeGPU)
+		ctx.RenderWithErr(ctx.Tr("repo.cloudbrain_bootfile_err"), tplGrampusTrainJobGPUNew, &form)
+		return
+	}
+
 	errStr := checkSpecialPool(ctx, "GPU")
 	if errStr != "" {
 		grampusTrainJobNewDataPrepare(ctx, grampus.ProcessorTypeGPU)
@@ -399,6 +407,14 @@ func GrampusTrainJobNpuCreate(ctx *context.Context, form auth.CreateGrampusTrain
 		return
 	}
 
+	bootFileExist, err := ctx.Repo.FileExists(bootFile, branchName)
+	if err != nil || !bootFileExist {
+		log.Error("Get bootfile error:", err, ctx.Data["MsgID"])
+		grampusTrainJobNewDataPrepare(ctx, grampus.ProcessorTypeNPU)
+		ctx.RenderWithErr(ctx.Tr("repo.cloudbrain_bootfile_err"), tplGrampusTrainJobNPUNew, &form)
+		return
+	}
+
 	errStr := checkSpecialPool(ctx, "NPU")
 	if errStr != "" {
 		grampusTrainJobNewDataPrepare(ctx, grampus.ProcessorTypeNPU)

routers/repo/modelarts.go

@@ -1089,7 +1089,7 @@ func TrainJobCreate(ctx *context.Context, form auth.CreateModelArtsTrainJobForm)
 	outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath + VersionOutputPath + "/"
 	logObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.LogPath + VersionOutputPath + "/"
 	// dataPath := "/" + setting.Bucket + "/" + setting.BasePath + path.Join(uuid[0:1], uuid[1:2]) + "/" + uuid + uuid + "/"
-	branch_name := form.BranchName
+	branchName := form.BranchName
 	isLatestVersion := modelarts.IsLatestVersion
 	FlavorName := form.FlavorName
 	VersionCount := modelarts.VersionCount
@@ -1117,6 +1117,14 @@ func TrainJobCreate(ctx *context.Context, form auth.CreateModelArtsTrainJobForm)
 		return
 	}
 
+	bootFileExist, err := ctx.Repo.FileExists(bootFile, branchName)
+	if err != nil || !bootFileExist {
+		log.Error("Get bootfile error:", err)
+		trainJobErrorNewDataPrepare(ctx, form)
+		ctx.RenderWithErr(ctx.Tr("repo.cloudbrain_bootfile_err"), tplModelArtsTrainJobNew, &form)
+		return
+	}
+
 	errStr := checkModelArtsSpecialPool(ctx, flavorCode, string(models.JobTypeTrain))
 	if errStr != "" {
 		trainJobErrorNewDataPrepare(ctx, form)
@@ -1148,9 +1156,9 @@ func TrainJobCreate(ctx *context.Context, form auth.CreateModelArtsTrainJobForm)
 	}
 
 	gitRepo, _ := git.OpenRepository(repo.RepoPath())
-	commitID, _ := gitRepo.GetBranchCommitID(branch_name)
+	commitID, _ := gitRepo.GetBranchCommitID(branchName)
 
-	if err := downloadCode(repo, codeLocalPath, branch_name); err != nil {
+	if err := downloadCode(repo, codeLocalPath, branchName); err != nil {
 		log.Error("downloadCode failed, server timed out: %s (%v)", repo.FullName(), err)
 		trainJobErrorNewDataPrepare(ctx, form)
 		ctx.RenderWithErr(ctx.Tr("cloudbrain.load_code_failed"), tplModelArtsTrainJobNew, &form)
@@ -1292,7 +1300,7 @@ func TrainJobCreate(ctx *context.Context, form auth.CreateModelArtsTrainJobForm)
 		Parameters:        param,
 		CommitID:          commitID,
 		IsLatestVersion:   isLatestVersion,
-		BranchName:        branch_name,
+		BranchName:        branchName,
 		Params:            form.Params,
 		FlavorName:        FlavorName,
 		EngineName:        EngineName,
@@ -1394,7 +1402,7 @@ func TrainJobCreateVersion(ctx *context.Context, form auth.CreateModelArtsTrainJ
 	outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath + VersionOutputPath + "/"
 	logObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.LogPath + VersionOutputPath + "/"
 	// dataPath := "/" + setting.Bucket + "/" + setting.BasePath + path.Join(uuid[0:1], uuid[1:2]) + "/" + uuid + uuid + "/"
-	branch_name := form.BranchName
+	branchName := form.BranchName
 	PreVersionName := form.VersionName
 	FlavorName := form.FlavorName
 	EngineName := form.EngineName
@@ -1414,6 +1422,14 @@ func TrainJobCreateVersion(ctx *context.Context, form auth.CreateModelArtsTrainJ
 		return
 	}
 
+	bootFileExist, err := ctx.Repo.FileExists(bootFile, branchName)
+	if err != nil || !bootFileExist {
+		log.Error("Get bootfile error:", err)
+		versionErrorDataPrepare(ctx, form)
+		ctx.RenderWithErr(ctx.Tr("repo.cloudbrain_bootfile_err"), tplModelArtsTrainJobVersionNew, &form)
+		return
+	}
+
 	errStr := checkModelArtsSpecialPool(ctx, flavorCode, string(models.JobTypeTrain))
 	if errStr != "" {
 		versionErrorDataPrepare(ctx, form)
@@ -1428,8 +1444,8 @@ func TrainJobCreateVersion(ctx *context.Context, form auth.CreateModelArtsTrainJ
 	}
 
 	gitRepo, _ := git.OpenRepository(repo.RepoPath())
-	commitID, _ := gitRepo.GetBranchCommitID(branch_name)
-	if err := downloadCode(repo, codeLocalPath, branch_name); err != nil {
+	commitID, _ := gitRepo.GetBranchCommitID(branchName)
+	if err := downloadCode(repo, codeLocalPath, branchName); err != nil {
 		log.Error("Failed git clone repo to local(！: %s (%v)", repo.FullName(), err)
 		versionErrorDataPrepare(ctx, form)
 		ctx.RenderWithErr(ctx.Tr("cloudbrain.load_code_failed"), tplModelArtsTrainJobVersionNew, &form)
@@ -1582,7 +1598,7 @@ func TrainJobCreateVersion(ctx *context.Context, form auth.CreateModelArtsTrainJ
 		Parameters:        param,
 		PreVersionId:      task.VersionID,
 		CommitID:          commitID,
-		BranchName:        branch_name,
+		BranchName:        branchName,
 		FlavorName:        FlavorName,
 		EngineName:        EngineName,
 		PreVersionName:    PreVersionName,
@@ -2025,7 +2041,7 @@ func InferenceJobCreate(ctx *context.Context, form auth.CreateModelArtsInference
 	resultObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.ResultPath + VersionOutputPath + "/"
 	logObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.LogPath + VersionOutputPath + "/"
 	dataPath := "/" + setting.Bucket + "/" + setting.BasePath + path.Join(uuid[0:1], uuid[1:2]) + "/" + uuid + uuid + "/"
-	branch_name := form.BranchName
+	branchName := form.BranchName
 	FlavorName := form.FlavorName
 	EngineName := form.EngineName
 	LabelName := form.LabelName
@@ -2060,6 +2076,14 @@ func InferenceJobCreate(ctx *context.Context, form auth.CreateModelArtsInference
 		return
 	}
 
+	bootFileExist, err := ctx.Repo.FileExists(bootFile, branchName)
+	if err != nil || !bootFileExist {
+		log.Error("Get bootfile error:", err)
+		inferenceJobErrorNewDataPrepare(ctx, form)
+		ctx.RenderWithErr(ctx.Tr("repo.cloudbrain_bootfile_err"), tplModelArtsInferenceJobNew, &form)
+		return
+	}
+
 	//Determine whether the task name of the task in the project is duplicated
 	tasks, err := models.GetCloudbrainsByDisplayJobName(repo.ID, string(models.JobTypeInference), displayJobName)
 	if err == nil {
@@ -2092,9 +2116,9 @@ func InferenceJobCreate(ctx *context.Context, form auth.CreateModelArtsInference
 	}
 
 	gitRepo, _ := git.OpenRepository(repo.RepoPath())
-	commitID, _ := gitRepo.GetBranchCommitID(branch_name)
+	commitID, _ := gitRepo.GetBranchCommitID(branchName)
 
-	if err := downloadCode(repo, codeLocalPath, branch_name); err != nil {
+	if err := downloadCode(repo, codeLocalPath, branchName); err != nil {
 		log.Error("Create task failed, server timed out: %s (%v)", repo.FullName(), err)
 		inferenceJobErrorNewDataPrepare(ctx, form)
 		ctx.RenderWithErr(ctx.Tr("cloudbrain.load_code_failed"), tplModelArtsInferenceJobNew, &form)
@@ -2178,7 +2202,7 @@ func InferenceJobCreate(ctx *context.Context, form auth.CreateModelArtsInference
 		Uuid:              uuid,
 		Parameters:        param, //modelarts train parameters
 		CommitID:          commitID,
-		BranchName:        branch_name,
+		BranchName:        branchName,
 		Params:            form.Params,
 		FlavorName:        FlavorName,
 		EngineName:        EngineName,