debug

modules/grampus/grampus.go

@@ -40,7 +40,7 @@ const (
 	ProcessorTypeNPU = "npu.huawei.com/NPU"
 	ProcessorTypeGPU = "nvidia.com/gpu"
 
-	CommandPrepareScript = "pwd;cd /tmp;wget https://git.openi.org.cn/lewis/script_for_grampus/archive/master.zip;unzip master.zip;cd script_for_grampus;chmod 777 sync_for_arm;"
+	CommandPrepareScript = "pwd;cd /tmp;mkdir output;mkdir code;mkdir dataset;wget https://git.openi.org.cn/lewis/script_for_grampus/archive/master.zip;unzip master.zip;cd script_for_grampus;chmod 777 sync_for_arm uploader_for_grampus;"
 	CodeArchiveName      = "master.zip"
 )
 

routers/repo/grampus.go

@@ -216,7 +216,6 @@ func grampusParamCheckCreateTrainJob(form auth.CreateGrampusTrainJobForm) error
 }
 
 func GrampusTrainJobNpuCreate(ctx *context.Context, form auth.CreateGrampusTrainJobForm) {
-	VersionOutputPath := modelarts.GetOutputPathByCount(modelarts.TotalVersionCount)
 	displayJobName := form.DisplayJobName
 	jobName := util.ConvertDisplayJobNameToJobName(displayJobName)
 	uuid := form.Attachment
@@ -298,20 +297,13 @@ func GrampusTrainJobNpuCreate(ctx *context.Context, form auth.CreateGrampusTrain
 	}
 
 	//todo: upload code (send to file_server todo this work?)
-	if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.OutputPath + VersionOutputPath + "/"); err != nil {
+	if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.OutputPath); err != nil {
 		log.Error("Failed to obsMkdir_output: %s (%v)", repo.FullName(), err)
 		grampusTrainJobNpuNewDataPrepare(ctx)
 		ctx.RenderWithErr("Failed to obsMkdir_output", tplGrampusTrainJobNPUNew, &form)
 		return
 	}
 
-	if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.LogPath + VersionOutputPath + "/"); err != nil {
-		log.Error("Failed to obsMkdir_log: %s (%v)", repo.FullName(), err)
-		grampusTrainJobNpuNewDataPrepare(ctx)
-		ctx.RenderWithErr("Failed to obsMkdir_log", tplGrampusTrainJobNPUNew, &form)
-		return
-	}
-
 	if err := uploadCodeToObs(codeLocalPath, jobName, ""); err != nil {
 		log.Error("Failed to uploadCodeToObs: %s (%v)", repo.FullName(), err)
 		grampusTrainJobNpuNewDataPrepare(ctx)
@@ -321,7 +313,7 @@ func GrampusTrainJobNpuCreate(ctx *context.Context, form auth.CreateGrampusTrain
 
 	//prepare command
 	//todo: download code, download dataset, unzip dataset, exec code, upload model
-	command, err := generateCommand(repo.Name, grampus.ProcessorTypeNPU, codeObsPath+cloudbrain.DefaultBranchName+".zip", dataObsPath+attachment.Name, bootFile, params, "", attachment.Name)
+	command, err := generateCommand(repo.Name, grampus.ProcessorTypeNPU, codeObsPath+cloudbrain.DefaultBranchName+".zip", dataObsPath+attachment.Name, bootFile, params, setting.CodePathPrefix+jobName+modelarts.OutputPath, attachment.Name)
 	log.Info(command)
 
 	commitID, _ := ctx.Repo.GitRepo.GetBranchCommitID(branchName)
@@ -545,7 +537,7 @@ func GrampusGetLog(ctx *context.Context) {
 	return
 }
 
-func generateCommand(repoName, processorType, codeObsPath, dataObsPath, bootFile, paramSrc, outputPath, datasetName string) (string, error) {
+func generateCommand(repoName, processorType, codeObsPath, dataObsPath, bootFile, paramSrc, outputObsPath, datasetName string) (string, error) {
 	var command string
 
 	command += grampus.CommandPrepareScript
@@ -558,7 +550,7 @@ func generateCommand(repoName, processorType, codeObsPath, dataObsPath, bootFile
 	}
 
 	//unzip code & dataset
-	commandUnzip := "cd dataset;unzip " + datasetName + ";cd ../code;unzip master.zip;"
+	commandUnzip := "cd /tmp/dataset;unzip " + datasetName + ";cd /tmp/code;unzip master.zip;"
 	command += commandUnzip
 
 	//exec code
@@ -581,10 +573,12 @@ func generateCommand(repoName, processorType, codeObsPath, dataObsPath, bootFile
 		}
 	}
 
-	commandCode := "cd " + repoName + ";python " + bootFile + paramCode
+	commandCode := "cd " + repoName + ";python " + bootFile + paramCode + ";"
 	command += commandCode
 
 	//upload models
+	commandUpload := "cd /tmp/script_for_grampus/;./uploader_for_grampus " + setting.Bucket + " " + outputObsPath + " " + "/tmp/output/;"
+	command += commandUpload
 
 	return command, nil
 }