提交代码

options/locale/locale_en-US.ini

@@ -1213,6 +1213,7 @@ modelarts.infer_job.select_model = Select Model
 modelarts.infer_job.boot_file_helper=The startup file is the entry file for your program execution and must end in.py.Such as  inference.py, main.py, example/inference.py, case/main.py.
 modelarts.infer_job.tooltip = The model has been deleted and cannot be viewed.
 modelarts.download_log=Download log file
+modelarts.no_node_right = The value of 'Amount of Compute Node' is wrong, you have no right to use the current value of 'Amount of Compute Node'.
 
 
 debug_task_not_created = Debug task has not been created

options/locale/locale_zh-CN.ini

@@ -1226,6 +1226,7 @@ modelarts.infer_job.select_model = 选择模型
 modelarts.infer_job.boot_file_helper=启动文件是您程序执行的入口文件，必须是以.py结尾的文件。比如inference.py、main.py、example/inference.py、case/main.py。
 modelarts.infer_job.tooltip = 该模型已删除，无法查看。
 modelarts.download_log=下载日志文件
+modelarts.no_node_right = 计算节点数的值配置错误，您没有权限使用当前配置的计算节点数。
 
 
 debug_task_not_created = 未创建过调试任务

routers/repo/modelarts.go

@@ -1130,6 +1130,13 @@ func TrainJobCreate(ctx *context.Context, form auth.CreateModelArtsTrainJobForm)
 	VersionCount := modelarts.VersionCountOne
 	EngineName := form.EngineName
 
+	errStr:=checkMultiNode(ctx.User.ID,form.WorkServerNumber)
+	if errStr!=""{
+		trainJobErrorNewDataPrepare(ctx, form)
+		ctx.RenderWithErr(ctx.Tr(errStr), tplModelArtsTrainJobNew, &form)
+		return
+	}
+
 	count, err := models.GetCloudbrainTrainJobCountByUserID(ctx.User.ID)
 	if err != nil {
 		log.Error("GetCloudbrainTrainJobCountByUserID failed:%v", err, ctx.Data["MsgID"])
@@ -1160,7 +1167,7 @@ func TrainJobCreate(ctx *context.Context, form auth.CreateModelArtsTrainJobForm)
 		return
 	}
 
-	errStr := checkModelArtsSpecialPool(ctx, flavorCode, string(models.JobTypeTrain))
+	errStr = checkModelArtsSpecialPool(ctx, flavorCode, string(models.JobTypeTrain))
 	if errStr != "" {
 		trainJobErrorNewDataPrepare(ctx, form)
 		ctx.RenderWithErr(ctx.Tr(errStr), tplModelArtsTrainJobNew, &form)
@@ -1364,6 +1371,48 @@ func TrainJobCreate(ctx *context.Context, form auth.CreateModelArtsTrainJobForm)
 	ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job")
 }
 
+func checkMultiNode(userId int64, serverNum int) string{
+	if serverNum==1{
+		return ""
+	}
+	modelarts.InitMultiNode()
+	var isServerNumValid=false
+	if modelarts.MultiNodeConfig != nil {
+		for _, info := range modelarts.MultiNodeConfig.Info {
+			if isInOrg, _ := models.IsOrganizationMemberByOrgName(info.Org, userId); isInOrg {
+				if isInNodes(info.Node,serverNum){
+					isServerNumValid=true
+					break
+				}
+
+			}
+		}
+	}
+	if isServerNumValid{
+		return ""
+	}else{
+		return "repo.modelarts.no_node_right"
+	}
+}
+func checkInferenceJobMultiNode(userId int64, serverNum int) string{
+	if serverNum==1{
+		return ""
+	}
+
+	return "repo.modelarts.no_node_right"
+
+}
+
+func isInNodes(nodes []int, num int) bool {
+	for _, node:=range nodes{
+		if node==num{
+			return true
+		}
+	}
+	return false
+
+}
+
 func getUserCommand(engineId int, req *modelarts.GenerateTrainJobReq) (string, string) {
 	userImageUrl := ""
 	userCommand := ""
@@ -1398,6 +1447,13 @@ func TrainJobCreateVersion(ctx *context.Context, form auth.CreateModelArtsTrainJ
 	ctx.Data["PageIsTrainJob"] = true
 	var jobID = ctx.Params(":jobid")
 
+	errStr:=checkMultiNode(ctx.User.ID,form.WorkServerNumber)
+	if errStr!=""{
+		versionErrorDataPrepare(ctx, form)
+		ctx.RenderWithErr(ctx.Tr(errStr), tplModelArtsTrainJobVersionNew, &form)
+		return
+	}
+
 	count, err := models.GetCloudbrainTrainJobCountByUserID(ctx.User.ID)
 	if err != nil {
 		log.Error("GetCloudbrainTrainJobCountByUserID failed:%v", err, ctx.Data["MsgID"])
@@ -1465,7 +1521,7 @@ func TrainJobCreateVersion(ctx *context.Context, form auth.CreateModelArtsTrainJ
 		return
 	}
 
-	errStr := checkModelArtsSpecialPool(ctx, flavorCode, string(models.JobTypeTrain))
+	errStr = checkModelArtsSpecialPool(ctx, flavorCode, string(models.JobTypeTrain))
 	if errStr != "" {
 		versionErrorDataPrepare(ctx, form)
 		ctx.RenderWithErr(ctx.Tr(errStr), tplModelArtsTrainJobVersionNew, &form)
@@ -2036,6 +2092,13 @@ func InferenceJobCreate(ctx *context.Context, form auth.CreateModelArtsInference
 	ckptUrl := "/" + form.TrainUrl + form.CkptName
 	log.Info("ckpt url:" + ckptUrl)
 
+	errStr:=checkInferenceJobMultiNode(ctx.User.ID,form.WorkServerNumber)
+	if errStr!=""{
+		inferenceJobErrorNewDataPrepare(ctx, form)
+		ctx.RenderWithErr(ctx.Tr(errStr), tplModelArtsInferenceJobNew, &form)
+		return
+	}
+
 	count, err := models.GetCloudbrainInferenceJobCountByUserID(ctx.User.ID)
 	if err != nil {
 		log.Error("GetCloudbrainInferenceJobCountByUserID failed:%v", err, ctx.Data["MsgID"])
@@ -2084,7 +2147,7 @@ func InferenceJobCreate(ctx *context.Context, form auth.CreateModelArtsInference
 		}
 	}
 
-	errStr := checkModelArtsSpecialPool(ctx, flavorCode, string(models.JobTypeInference))
+	errStr = checkModelArtsSpecialPool(ctx, flavorCode, string(models.JobTypeInference))
 	if errStr != "" {
 		inferenceJobErrorNewDataPrepare(ctx, form)
 		ctx.RenderWithErr(ctx.Tr(errStr), tplModelArtsInferenceJobNew, &form)