Merge pull request 'updated createdeploytask logics' (#299) from tzwang/pcm-coordinator:master into master

Former-commit-id: 1e098dc4bca0bcc08c005b260141f5255416a8d3

desc/inference/inference.api

@@ -200,7 +200,7 @@ type (
 
      AdapterAvail {
           AdapterId string `json:"adapterId"`
-          AdapterName string `json:"taskName"`
+          AdapterName string `json:"adapterName"`
           Clusters []*ClusterAvail `json:"clusters"`
      }
 

internal/logic/inference/createdeploytasklogic.go

@@ -95,7 +95,7 @@ func (l *CreateDeployTaskLogic) createDeployInstance(taskId int64, adapterId str
 		return err
 	}
 
-	_, err = l.svcCtx.Scheduler.AiStorages.SaveInferDeployInstance(taskId, ins.InstanceId, ins.InstanceName, aid, adapterName, cid, clusterName, ins.ModelName, ins.ModelType, ins.InferCard)
+	_, err = l.svcCtx.Scheduler.AiStorages.SaveInferDeployInstance(taskId, ins.InstanceId, ins.InstanceName, aid, adapterName, cid, clusterName, ins.ModelName, ins.ModelType, ins.InferCard, ins.ClusterType)
 	if err != nil {
 		return err
 	}

internal/logic/inference/deployinstancelistlogic.go

@@ -52,7 +52,10 @@ func (l *DeployInstanceListLogic) DeployInstanceList(req *types.DeployInstanceLi
 		return nil, errors.New(err.Error())
 	}
 
-	deployTasks := l.GenerateDeployTasks(tasklist)
+	deployTasks, err := l.GenerateDeployTasks(tasklist)
+	if err != nil {
+		return nil, errors.New(err.Error())
+	}
 	slices := make([][]*models.AiInferDeployInstance, len(deployTasks))
 	for i := 0; i < len(deployTasks); i++ {
 		slices[i] = deployTasks[i].Instances
@@ -83,12 +86,20 @@ func (l *DeployInstanceListLogic) DeployInstanceList(req *types.DeployInstanceLi
 	return
 }
 
-func (l *DeployInstanceListLogic) GenerateDeployTasks(tasklist []*models.AiDeployInstanceTask) []*DeployTask {
+func (l *DeployInstanceListLogic) GenerateDeployTasks(tasklist []*models.AiDeployInstanceTask) ([]*DeployTask, error) {
 	var tasks []*DeployTask
 	for _, t := range tasklist {
 		list, err := l.svcCtx.Scheduler.AiStorages.GetInstanceListByDeployTaskId(t.Id)
 		if err != nil {
 			logx.Errorf("db GetInstanceListByDeployTaskId error")
+			return nil, errors.New(err.Error())
+		}
+		if len(list) == 0 {
+			err := l.svcCtx.Scheduler.AiStorages.DeleteDeployTaskById(t.Id)
+			if err != nil {
+				logx.Errorf("db DeleteByDeployTaskId error")
+				return nil, errors.New(err.Error())
+			}
 			continue
 		}
 		deployTask := &DeployTask{
@@ -99,7 +110,7 @@ func (l *DeployInstanceListLogic) GenerateDeployTasks(tasklist []*models.AiDeplo
 		}
 		tasks = append(tasks, deployTask)
 	}
-	return tasks
+	return tasks, nil
 }
 
 type DeployTask struct {

internal/scheduler/database/aiStorage.go

@@ -374,7 +374,7 @@ func (s *AiStorage) AddNoticeInfo(adapterId string, adapterName string, clusterI
 }
 
 func (s *AiStorage) SaveInferDeployInstance(taskId int64, instanceId string, instanceName string, adapterId int64,
-	adapterName string, clusterId int64, clusterName string, modelName string, modelType string, inferCard string) (int64, error) {
+	adapterName string, clusterId int64, clusterName string, modelName string, modelType string, inferCard string, clusterType string) (int64, error) {
 	startTime := time.Now().Format(time.RFC3339)
 	// 构建主任务结构体
 	insModel := models.AiInferDeployInstance{
@@ -388,7 +388,8 @@ func (s *AiStorage) SaveInferDeployInstance(taskId int64, instanceId string, ins
 		ModelName:            modelName,
 		ModelType:            modelType,
 		InferCard:            inferCard,
-		Status:               constants.Stopped,
+		ClusterType:          clusterType,
+		Status:               constants.Deploying,
 		CreateTime:           startTime,
 		UpdateTime:           startTime,
 	}
@@ -464,6 +465,15 @@ func (s *AiStorage) UpdateDeployTask(task *models.AiDeployInstanceTask, needUpda
 	return nil
 }
 
+func (s *AiStorage) DeleteDeployTaskById(id int64) error {
+	tx := s.DbEngin.Delete(&models.AiDeployInstanceTask{}, id)
+	if tx.Error != nil {
+		logx.Errorf(tx.Error.Error())
+		return tx.Error
+	}
+	return nil
+}
+
 func (s *AiStorage) UpdateDeployTaskById(id int64) error {
 	task, err := s.GetDeployTaskById(id)
 	if err != nil {

internal/storeLink/octopus.go

@@ -1158,10 +1158,14 @@ func (o *OctopusLink) GetInferDeployInstance(ctx context.Context, id string) (*i
 	url := strings.Replace(resp.Payload.Notebook.Tasks[0].Url, FORWARD_SLASH, "", -1)
 	inferUrl := DOMAIN + url
 
+	var modelType string
+	var modelName string
 	var card string
 	if resp.Payload.Notebook.Desc != "" {
 		str := strings.Split(resp.Payload.Notebook.Desc, FORWARD_SLASH)
 		if len(str) == 3 {
+			modelType = str[0]
+			modelName = str[1]
 			card = str[2]
 		}
 	}
@@ -1171,6 +1175,8 @@ func (o *OctopusLink) GetInferDeployInstance(ctx context.Context, id string) (*i
 	ins.ClusterName = o.platform
 	ins.Status = resp.Payload.Notebook.Status
 	ins.ClusterType = TYPE_OCTOPUS
+	ins.ModelType = modelType
+	ins.ModelName = modelName
 	ins.InferUrl = inferUrl
 	ins.InferCard = card
 

internal/types/types.go

@@ -6096,7 +6096,7 @@ type GetAdaptersByModelResp struct {
 
 type AdapterAvail struct {
 	AdapterId   string          `json:"adapterId"`
-	AdapterName string          `json:"taskName"`
+	AdapterName string          `json:"adapterName"`
 	Clusters    []*ClusterAvail `json:"clusters"`
 }