Merge pull request 'modify ai replicas' (#178) from master-wq into master

Former-commit-id: 1fe43cd8eff3a1986305a3d02fad93df149ace9e

api/desc/schedule/pcm-schedule.api

@@ -44,6 +44,7 @@ type (
         Params []string `json:"params,optional"`
         Envs   []string `json:"envs,optional"`
         Cmd string `json:"cmd,optional"`
+        Replica     int32     `json:"replicas"`
     }
 
     AiResourceTypesResp {

api/internal/logic/core/commitvmtasklogic.go

@@ -110,7 +110,6 @@ func (l *CommitVmTaskLogic) CommitVmTask(req *types.CommitVmTaskReq) (resp *type
 	                    AND vf.public_flavor_id = ?
 	                    AND vi.public_image_id = ?
 	                    AND vn.public_network_id = ?`
-				// err2 := l.svcCtx.DbEngin.Raw(sql, clusterId, req.FlavorRef, req.ImageRef, req.Uuid).Scan(&taskVm).Error
 				txVm := l.svcCtx.DbEngin.Raw(sql, clusterId, req.FlavorRef, req.ImageRef, req.Uuid).Scan(&VmObject)
 				if txVm.Error != nil {
 					logx.Error(err)
@@ -118,7 +117,6 @@ func (l *CommitVmTaskLogic) CommitVmTask(req *types.CommitVmTaskReq) (resp *type
 				}
 				if err2 != nil {
 					logx.Errorf("CommitGeneralTask() => sql execution error: %v", err)
-					//return errors.Errorf("the cluster does not match the drive resources. Check the data"), nil
 				}
 				taskVm.Name = req.VmName
 				taskVm.Status = "Saved"

api/internal/logic/schedule/schedulesubmitlogic.go

@@ -33,8 +33,8 @@ func (l *ScheduleSubmitLogic) ScheduleSubmit(req *types.ScheduleReq) (resp *type
 		AdapterId:             req.AiOption.AdapterId,
 		TaskName:              req.AiOption.TaskName,
 		ResourceType:          req.AiOption.ResourceType,
+		Replica:               req.AiOption.Replica,
 		ComputeCard:           req.AiOption.ComputeCard,
-		Replica:               1,
 		Tops:                  req.AiOption.Tops,
 		TaskType:              req.AiOption.TaskType,
 		DatasetsName:          req.AiOption.Datasets,

api/internal/scheduler/database/aiStorage.go

@@ -120,7 +120,7 @@ func (s *AiStorage) SaveAiTask(taskId int64, option *option.AiOption, clusterId
 		ClusterId:   cId,
 		ClusterName: clusterName,
 		Name:        option.TaskName,
-		Replica:     option.Replica,
+		Replica:    int64(option.Replica),
 		JobId:       jobId,
 		TaskType:    option.TaskType,
 		Strategy:    option.StrategyName,

api/internal/scheduler/schedulers/aiScheduler.go

@@ -105,7 +105,7 @@ func (as *AiScheduler) PickOptimalStrategy() (strategy.Strategy, error) {
 		return strategy, nil
 	case strategy.STATIC_WEIGHT:
 		//todo resources should match cluster StaticWeightMap
-		strategy := strategy.NewStaticWeightStrategy(as.option.ClusterToStaticWeight, 1)
+		strategy := strategy.NewStaticWeightStrategy(as.option.ClusterToStaticWeight, as.option.Replica)
 		return strategy, nil
 	}
 

api/internal/scheduler/schedulers/option/aiOption.go

@@ -4,7 +4,7 @@ type AiOption struct {
 	AdapterId             string
 	ClusterIds            []string
 	TaskName              string
-	Replica               int64
+	Replica               int32
 	ResourceType          string // cpu/gpu/compute card
 	CpuCoreNum            int64
 	TaskType              string // pytorch/tensorflow/mindspore

api/internal/types/types.go

@@ -5647,6 +5647,7 @@ type AiOption struct {
 	Params          []string         `json:"params,optional"`
 	Envs            []string         `json:"envs,optional"`
 	Cmd             string           `json:"cmd,optional"`
+	Replica         int32            `json:"replicas"`
 }
 
 type AiResourceTypesResp struct {