提交代码。

Signed-off-by: zouap <zouap@pcl.ac.cn>

models/cloudbrain.go

@@ -2006,9 +2006,9 @@ func GetStoppedJobWithNoStartTimeEndTime() ([]*Cloudbrain, error) {
 	return cloudbrains, x.SQL("select * from cloudbrain where status in (?,?,?,?,?,?,?) and (start_time is null or end_time is null) limit 100", ModelArtsTrainJobCompleted, ModelArtsTrainJobFailed, ModelArtsTrainJobKilled, ModelArtsStopped, JobStopped, JobFailed, JobSucceeded).Find(&cloudbrains)
 }
 
-func GetRunningModelSafetyTestTask() ([]*Cloudbrain, error) {
+func GetModelSafetyTestTask() ([]*Cloudbrain, error) {
 	cloudbrains := make([]*Cloudbrain, 0)
-	sess := x.Where("job_type = ? and status=?", string(JobTypeModelSafety), JobRunning)
+	sess := x.Where("job_type = ?", string(JobTypeModelSafety))
 	err := sess.Find(&cloudbrains)
 	return cloudbrains, err
 }

routers/repo/aisafety.go

@@ -63,62 +63,87 @@ func CloudBrainAiSafetyCreateTest(ctx *context.Context) {
 
 func GetAiSafetyTask(ctx *context.Context) {
 	var ID = ctx.Params(":jobid")
-	getAiSafetyTaskStatusFromCloudbrain(ID)
-}
-
-func getAiSafetyTaskStatusFromCloudbrain(ID string) {
 	job, err := models.GetCloudbrainByJobIDWithDeleted(ID)
 	if err != nil {
 		log.Error("GetCloudbrainByJobID failed:" + err.Error())
 		return
 	}
+	syncAiSafetyTaskStatus(job)
+}
+
+func syncAiSafetyTaskStatus(job *models.Cloudbrain) {
 	if job.Type == models.TypeCloudBrainTwo {
 
 	} else if job.Type == models.TypeCloudBrainOne {
 		if isTaskNotFinished(job.Status) {
-			log.Info("The task not finished,name=" + job.DisplayJobName)
-			jobResult, err := cloudbrain.GetJob(job.JobID)
-
-			result, err := models.ConvertToJobResultPayload(jobResult.Payload)
-			if err != nil {
-				log.Error("ConvertToJobResultPayload failed:", err)
-				return
-			}
-			job.Status = result.JobStatus.State
-			if result.JobStatus.State != string(models.JobWaiting) && result.JobStatus.State != string(models.JobFailed) {
-				taskRoles := result.TaskRoles
-				taskRes, _ := models.ConvertToTaskPod(taskRoles[cloudbrain.SubTaskName].(map[string]interface{}))
-				job.Status = taskRes.TaskStatuses[0].State
-			}
-
-			if result.JobStatus.State != string(models.JobSucceeded) {
-				err = models.UpdateJob(job)
-				if err != nil {
-					log.Error("UpdateJob failed:", err)
-				}
+			queryTaskStatusFromCloudbrain(job)
+		} else {
+			if job.Status == string(models.ModelSafetyTesting) {
+				queryTaskStatusFromModelSafetyTestServer(job)
 			} else {
-				//
-				job.Status = string(models.ModelSafetyTesting)
-				err = models.UpdateJob(job)
-				if err != nil {
-					log.Error("UpdateJob failed:", err)
-				}
-				//send msg to beihang
-				sendGpuInferenceResultToTest(job)
+				log.Info("The job is finished. status=" + job.Status)
 			}
+		}
+	}
+}
 
+func TimerHandleModelSafetyTestTask() {
+	tasks, err := models.GetModelSafetyTestTask()
+	if err == nil {
+		if tasks != nil && len(tasks) > 0 {
+			for _, job := range tasks {
+				syncAiSafetyTaskStatus(job)
+			}
 		} else {
-			if job.Status == string(models.ModelSafetyTesting) {
-				//
-				result, err := aisafety.GetTaskStatus(job.PreVersionName)
-				if err == nil {
-					if result["code"] != nil {
+			log.Info("query running model safety test task 0.")
+		}
+	} else {
+		log.Info("query running model safety test task err." + err.Error())
+	}
+}
 
-					}
-				}
-			}
+func queryTaskStatusFromCloudbrain(job *models.Cloudbrain) {
+
+	log.Info("The task not finished,name=" + job.DisplayJobName)
+	jobResult, err := cloudbrain.GetJob(job.JobID)
+
+	result, err := models.ConvertToJobResultPayload(jobResult.Payload)
+	if err != nil {
+		log.Error("ConvertToJobResultPayload failed:", err)
+		return
+	}
+	job.Status = result.JobStatus.State
+	if result.JobStatus.State != string(models.JobWaiting) && result.JobStatus.State != string(models.JobFailed) {
+		taskRoles := result.TaskRoles
+		taskRes, _ := models.ConvertToTaskPod(taskRoles[cloudbrain.SubTaskName].(map[string]interface{}))
+		job.Status = taskRes.TaskStatuses[0].State
+	}
+
+	if result.JobStatus.State != string(models.JobSucceeded) {
+		err = models.UpdateJob(job)
+		if err != nil {
+			log.Error("UpdateJob failed:", err)
+		}
+	} else {
+		//
+		job.Status = string(models.ModelSafetyTesting)
+		err = models.UpdateJob(job)
+		if err != nil {
+			log.Error("UpdateJob failed:", err)
 		}
+		//send msg to beihang
+		sendGpuInferenceResultToTest(job)
+	}
+}
 
+func queryTaskStatusFromModelSafetyTestServer(job *models.Cloudbrain) {
+	result, err := aisafety.GetTaskStatus(job.PreVersionName)
+	if err == nil {
+		if result["code"] != nil {
+			if result["code"].(string) == "0" {
+
+			}
+		}
 	}
 }
 
@@ -272,6 +297,15 @@ func AiSafetyCreateForPost(ctx *context.Context) {
 			return
 		}
 	}
+	BootFile := ctx.Query("BootFile")
+	bootFileExist, err := ctx.Repo.FileExists(BootFile, cloudbrain.DefaultBranchName)
+	if err != nil || !bootFileExist {
+		log.Error("Get bootfile error:", err)
+		modelSafetyNewDataPrepare(ctx)
+		ctx.RenderWithErr(ctx.Tr("repo.cloudbrain_bootfile_err"), tplModelArtsInferenceJobNew, nil)
+		return
+	}
+
 	if taskType == models.TypeCloudBrainTwo {
 		createForNPU(ctx, jobName)
 	} else if taskType == models.TypeCloudBrainOne {
@@ -280,35 +314,6 @@ func AiSafetyCreateForPost(ctx *context.Context) {
 	//ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/cloudbrain/benchmark")
 }
 
-func TimerHandleModelSafetyTestTask() {
-	tasks, err := models.GetRunningModelSafetyTestTask()
-	if err == nil {
-		if tasks != nil && len(tasks) > 0 {
-			for _, task := range tasks {
-				if task.Type == models.TypeCloudBrainOne {
-					if task.Status == string(models.JobRunning) {
-						//调用云脑任务状态接口
-
-					} else if task.Status == string(models.ModelSafetyTesting) {
-						//调用模型评测查询状态接口
-
-					}
-				} else if task.Type == models.TypeCloudBrainTwo {
-					if task.Status == string(models.ModelArtsRunning) {
-
-					} else if task.Status == string(models.ModelArtsTrainJobCompleted) {
-
-					}
-				}
-			}
-		} else {
-			log.Info("query running model safety test task 0.")
-		}
-	} else {
-		log.Info("query running model safety test task err." + err.Error())
-	}
-}
-
 func createForNPU(ctx *context.Context, jobName string) {
 
 }
@@ -349,19 +354,6 @@ func createForGPU(ctx *context.Context, jobName string) {
 		return
 	}
 
-	if _, err := os.Stat(codePath + "/" + BootFile); err != nil {
-		if os.IsNotExist(err) {
-			// file does not exist
-			log.Error(BootFile+" does not exist, %v", err, ctx.Data["MsgID"])
-			modelSafetyNewDataPrepare(ctx)
-			ctx.RenderWithErr(BootFile+" does not exist", tplCloudBrainModelSafetyNew, nil)
-		} else {
-			log.Error("Stat failed, %v", err, ctx.Data["MsgID"])
-			modelSafetyNewDataPrepare(ctx)
-			ctx.RenderWithErr("system error", tplCloudBrainModelSafetyNew, nil)
-		}
-		return
-	}
 	err = uploadCodeToMinio(codePath+"/", jobName, cloudbrain.CodeMountPath+"/")
 	if err != nil {
 		log.Error("uploadCodeToMinio failed, %v", err, ctx.Data["MsgID"])