#2701

update

models/cloudbrain.go

@@ -1947,11 +1947,18 @@ func RestartCloudbrain(old *Cloudbrain, new *Cloudbrain) (err error) {
 		return err
 	}
 
-	if _, err = sess.NoAutoTime().Insert(new); err != nil {
+	if _, err = sess.NoAutoTime().InsertOne(new); err != nil {
 		sess.Rollback()
 		return err
 	}
 
+	if new.Spec != nil {
+		if _, err = sess.Insert(NewCloudBrainSpec(new.ID, *new.Spec)); err != nil {
+			sess.Rollback()
+			return err
+		}
+	}
+
 	if err = sess.Commit(); err != nil {
 		return err
 	}

modules/cloudbrain/cloudbrain.go

@@ -409,25 +409,7 @@ func GetWaitingCloudbrainCount(cloudbrainType int, computeResource string, jobTy
 func RestartTask(ctx *context.Context, task *models.Cloudbrain, newID *string) error {
 	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
-		}
-	}
-
-	//如果没有匹配到spec信息，尝试从专属资源池获取
-	if resourceSpec == nil && SpecialPools != nil {
-		resourceSpec = geMatchResourceSpec(task.JobType, task.GpuQueue, task.ResourceSpecId)
-	}
-
-	if resourceSpec == nil {
-		log.Error("no such resourceSpecId(%d)", task.ResourceSpecId, ctx.Data["MsgID"])
-		return errors.New("no such resourceSpec")
-	}
+	spec := task.Spec
 	var datasetInfos map[string]models.DatasetInfo
 	if task.Uuid != "" {
 		var err error
@@ -509,10 +491,10 @@ func RestartTask(ctx *context.Context, task *models.Cloudbrain, newID *string) e
 				TaskNumber:            1,
 				MinSucceededTaskCount: 1,
 				MinFailedTaskCount:    1,
-				CPUNumber:             resourceSpec.CpuNum,
-				GPUNumber:             resourceSpec.GpuNum,
-				MemoryMB:              resourceSpec.MemMiB,
-				ShmMB:                 resourceSpec.ShareMemMiB,
+				CPUNumber:             spec.CpuCores,
+				GPUNumber:             spec.AccCardsNum,
+				MemoryMB:              int(spec.MemGiB * 1024),
+				ShmMB:                 int(spec.ShareMemGiB * 1024),
 				Command:               GetCloudbrainDebugCommand(), //Command,
 				NeedIBDevice:          false,
 				IsMainRole:            false,
@@ -550,6 +532,7 @@ func RestartTask(ctx *context.Context, task *models.Cloudbrain, newID *string) e
 		CreatedUnix:     createTime,
 		UpdatedUnix:     createTime,
 		BranchName:      task.BranchName,
+		Spec:            spec,
 	}
 
 	err = models.RestartCloudbrain(task, newTask)

routers/repo/cloudbrain.go

@@ -560,34 +560,25 @@ func CloudBrainRestart(ctx *context.Context) {
 			break
 		}
 
-		var hasSameResource bool
-		if gpuInfos == nil {
-			json.Unmarshal([]byte(setting.GpuTypes), &gpuInfos)
-		}
-		for _, resourceType := range gpuInfos.GpuInfo {
-			if resourceType.Queue == task.GpuQueue {
-				hasSameResource = true
-				break
-			}
-		}
-		if !hasSameResource && cloudbrain.SpecialPools != nil {
-
-			for _, specialPool := range cloudbrain.SpecialPools.Pools {
-				cloudbrain.IsElementExist(specialPool.JobType, string(models.JobTypeDebug))
-				for _, pool := range specialPool.Pool {
-					if pool.Queue == task.GpuQueue {
-						hasSameResource = true
-					}
-				}
-			}
+		specOld, err := resource.GetCloudbrainSpec(task.ID)
+		if err != nil {
+			log.Error("CloudBrainRestart GetCloudbrainSpec error.task.id = %d", task.ID)
+			resultCode = "-1"
+			errorMsg = "Resource specification not support any more"
+			break
 		}
-
-		if !hasSameResource {
-			log.Error("has no same resource, can not restart", ctx.Data["MsgID"])
+		spec, err := resource.GetAndCheckSpec(ctx.User.ID, specOld.ID, models.FindSpecsOptions{
+			JobType:         models.JobType(task.JobType),
+			ComputeResource: models.GPU,
+			Cluster:         models.OpenICluster,
+			AiCenterCode:    models.AICenterOfCloudBrainOne})
+		if err != nil || spec == nil {
+			log.Error("CloudBrainRestart GetAndCheckSpec error.task.id = %d", task.ID)
 			resultCode = "-1"
-			errorMsg = "the job's version is too old and can not be restarted"
+			errorMsg = "Resource specification not support any more"
 			break
 		}
+		task.Spec = spec
 
 		count, err := models.GetCloudbrainCountByUserID(ctx.User.ID, string(models.JobTypeDebug))
 		if err != nil {