updated deployinstance status files

internal/cron/cron.go

@@ -16,7 +16,8 @@ package cron
 
 import (
 	"github.com/zeromicro/go-zero/core/logx"
-	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/updater"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/utils/stat"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/utils/status"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
 )
 
@@ -28,8 +29,8 @@ func AddCronGroup(svc *svc.ServiceContext) {
 			logx.Errorf(err.Error())
 			return
 		}
-		updater.UpdateTaskStatus(svc, list)
+		status.UpdateTaskStatus(svc, list)
-		updater.UpdateAiTaskStatus(svc, list)
+		status.UpdateAiTaskStatus(svc, list)
 	})
 
 	svc.Cron.AddFunc("*/5  *  *  *  *  ?", func() {
@@ -42,6 +43,6 @@ func AddCronGroup(svc *svc.ServiceContext) {
 			logx.Errorf(err.Error())
 			return
 		}
-		updater.UpdateClusterResources(svc, adapterList)
+		stat.UpdateClusterResources(svc, adapterList)
 	})
 }

internal/logic/ai/getcenteroverviewlogic.go

@@ -3,7 +3,7 @@ package ai
 import (
 	"context"
 	"github.com/zeromicro/go-zero/core/logx"
-	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/updater"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/utils/stat"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
 )
@@ -37,7 +37,7 @@ func (l *GetCenterOverviewLogic) GetCenterOverview() (resp *types.CenterOverview
 	centerNum = int32(len(adapterList))
 	resp.CenterNum = centerNum
 
-	go updater.UpdateClusterResources(l.svcCtx, adapterList)
+	go stat.UpdateClusterResources(l.svcCtx, adapterList)
 
 	for _, adapter := range adapterList {
 		taskList, err := l.svcCtx.Scheduler.AiStorages.GetAiTasksByAdapterId(adapter.Id)

internal/logic/ai/getcentertasklistlogic.go

@@ -3,7 +3,7 @@ package ai
 import (
 	"context"
 	"github.com/zeromicro/go-zero/core/logx"
-	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/updater"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/utils/status"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/constants"
@@ -32,7 +32,7 @@ func (l *GetCenterTaskListLogic) GetCenterTaskList() (resp *types.CenterTaskList
 		return nil, err
 	}
 
-	go updater.UpdateTrainingTaskStatus(l.svcCtx, adapterList)
+	go status.UpdateTrainingTaskStatus(l.svcCtx, adapterList)
 
 	for _, adapter := range adapterList {
 		taskList, err := l.svcCtx.Scheduler.AiStorages.GetAiTasksByAdapterId(adapter.Id)

internal/logic/core/pagelisttasklogic.go

@@ -2,7 +2,7 @@ package core
 
 import (
 	"context"
-	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/updater"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/utils/status"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/repository/result"
@@ -52,8 +52,8 @@ func (l *PageListTaskLogic) PageListTask(req *types.PageTaskReq) (resp *types.Pa
 	}
 
 	// 更新智算任务状态
-	go updater.UpdateTaskStatus(l.svcCtx, list)
+	go status.UpdateTaskStatus(l.svcCtx, list)
-	go updater.UpdateAiTaskStatus(l.svcCtx, list)
+	go status.UpdateAiTaskStatus(l.svcCtx, list)
 
 	for _, model := range list {
 		if model.StartTime != "" && model.EndTime == "" {

internal/logic/inference/deployinstancelistlogic.go

@@ -5,7 +5,7 @@ import (
 	"errors"
 	"github.com/zeromicro/go-zero/core/logx"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/common"
-	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/updater"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/utils/status"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/models"
@@ -71,7 +71,7 @@ func (l *DeployInstanceListLogic) DeployInstanceList(req *types.DeployInstanceLi
 	list := common.ConcatMultipleSlices(slices)
 
 	if len(list) != 0 {
-		go updater.UpdateDeployInstanceStatusBatch(l.svcCtx, list)
+		go status.UpdateDeployInstanceStatusBatch(l.svcCtx, list, true)
 
 		ins := list[0]
 		for i := range list {
@@ -82,8 +82,8 @@ func (l *DeployInstanceListLogic) DeployInstanceList(req *types.DeployInstanceLi
 			}
 		}
 
-		go updater.UpdateDeployInstanceStatus(l.svcCtx, ins, true)
+		go status.UpdateDeployInstanceStatus(l.svcCtx, ins, true)
-		go updater.UpdateDeployTaskStatus(l.svcCtx)
+		go status.UpdateDeployTaskStatus(l.svcCtx)
 	}
 
 	resp.List = &deployTasks

internal/logic/inference/startallbydeploytaskidlogic.go

@@ -5,8 +5,7 @@ import (
 	"errors"
 	"fmt"
 	"github.com/zeromicro/go-zero/core/logx"
-	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/inference"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/utils/status"
-	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/storeLink"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/models"
@@ -84,7 +83,7 @@ func (l *StartAllByDeployTaskIdLogic) startAll(list []*models.AiInferDeployInsta
 				<-buf
 				return
 			}
-			if checkStopStatus(in) {
+			if status.CheckStopStatus(in) {
 				success := l.svcCtx.Scheduler.AiService.InferenceAdapterMap[strconv.FormatInt(ins.AdapterId, 10)][strconv.FormatInt(ins.ClusterId, 10)].StartInferDeployInstance(l.ctx, ins.InstanceId)
 				if !success {
 					e := struct {
@@ -136,31 +135,3 @@ func (l *StartAllByDeployTaskIdLogic) startAll(list []*models.AiInferDeployInsta
 
 	return nil
 }
-
-func checkStopStatus(in *inference.DeployInstance) bool {
-	switch in.ClusterType {
-	case storeLink.TYPE_OCTOPUS:
-		switch in.Status {
-		case "stopped":
-			return true
-		default:
-			return false
-		}
-	case storeLink.TYPE_MODELARTS:
-		switch in.Status {
-		case "stopped":
-			return true
-		default:
-			return false
-		}
-	case storeLink.TYPE_SHUGUANGAI:
-		switch in.Status {
-		case "Terminated":
-			return true
-		default:
-			return false
-		}
-	default:
-		return false
-	}
-}

internal/logic/inference/startdeployinstancelistlogic.go

@@ -4,7 +4,7 @@ import (
 	"context"
 	"errors"
 	"github.com/zeromicro/go-zero/core/logx"
-	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/updater"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/utils/status"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
 	"strconv"
@@ -33,12 +33,16 @@ func (l *StartDeployInstanceListLogic) StartDeployInstanceList(req *types.StartD
 		return nil, err
 	}
 
+	in, err := l.svcCtx.Scheduler.AiService.InferenceAdapterMap[strconv.FormatInt(ins.AdapterId, 10)][strconv.FormatInt(ins.ClusterId, 10)].GetInferDeployInstance(l.ctx, ins.InstanceId)
+
+	if status.CheckStopStatus(in) {
 		success := l.svcCtx.Scheduler.AiService.InferenceAdapterMap[req.AdapterId][req.ClusterId].StartInferDeployInstance(l.ctx, req.InstanceId)
 		if !success {
 			return nil, errors.New("start instance failed")
 		}
+	}
 
-	go updater.UpdateDeployInstanceStatus(l.svcCtx, ins, true)
+	go status.UpdateDeployInstanceStatus(l.svcCtx, ins, true)
 
 	return resp, nil
 }

internal/logic/inference/stopallbydeploytaskidlogic.go

@@ -4,8 +4,7 @@ import (
 	"context"
 	"errors"
 	"fmt"
-	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/inference"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/utils/status"
-	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/storeLink"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/models"
@@ -85,7 +84,7 @@ func (l *StopAllByDeployTaskIdLogic) stopAll(list []*models.AiInferDeployInstanc
 				<-buf
 				return
 			}
-			if checkStatus(in) {
+			if status.CheckRunningStatus(in) {
 				success := l.svcCtx.Scheduler.AiService.InferenceAdapterMap[strconv.FormatInt(ins.AdapterId, 10)][strconv.FormatInt(ins.ClusterId, 10)].StopInferDeployInstance(l.ctx, ins.InstanceId)
 				if !success {
 					e := struct {
@@ -137,31 +136,3 @@ func (l *StopAllByDeployTaskIdLogic) stopAll(list []*models.AiInferDeployInstanc
 
 	return nil
 }
-
-func checkStatus(in *inference.DeployInstance) bool {
-	switch in.ClusterType {
-	case storeLink.TYPE_OCTOPUS:
-		switch in.Status {
-		case "running":
-			return true
-		default:
-			return false
-		}
-	case storeLink.TYPE_MODELARTS:
-		switch in.Status {
-		case "running":
-			return true
-		default:
-			return false
-		}
-	case storeLink.TYPE_SHUGUANGAI:
-		switch in.Status {
-		case "Running":
-			return true
-		default:
-			return false
-		}
-	default:
-		return false
-	}
-}

internal/logic/inference/stopdeployinstancelogic.go

@@ -4,7 +4,7 @@ import (
 	"context"
 	"errors"
 	"github.com/zeromicro/go-zero/core/logx"
-	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/updater"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/utils/status"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
 	"strconv"
@@ -33,12 +33,16 @@ func (l *StopDeployInstanceLogic) StopDeployInstance(req *types.StopDeployInstan
 		return nil, err
 	}
 
+	in, err := l.svcCtx.Scheduler.AiService.InferenceAdapterMap[strconv.FormatInt(ins.AdapterId, 10)][strconv.FormatInt(ins.ClusterId, 10)].GetInferDeployInstance(l.ctx, ins.InstanceId)
+
+	if status.CheckRunningStatus(in) {
 		success := l.svcCtx.Scheduler.AiService.InferenceAdapterMap[req.AdapterId][req.ClusterId].StopInferDeployInstance(l.ctx, req.InstanceId)
 		if !success {
 			return nil, errors.New("stop instance failed")
 		}
+	}
 
-	go updater.UpdateDeployInstanceStatus(l.svcCtx, ins, true)
+	go status.UpdateDeployInstanceStatus(l.svcCtx, ins, true)
 
 	return resp, nil
 }

internal/scheduler/service/utils/stat/clusterResources.go

@@ -1,4 +1,4 @@
-package updater
+package stat
 
 import (
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"

internal/scheduler/service/utils/status/deployInstance.go

@@ -1,6 +1,7 @@
-package updater
+package status
 
 import (
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/inference"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/storeLink"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/constants"
@@ -10,14 +11,17 @@ import (
 	"time"
 )
 
-func UpdateDeployInstanceStatusBatch(svc *svc.ServiceContext, insList []*models.AiInferDeployInstance) {
+func UpdateDeployInstanceStatusBatch(svc *svc.ServiceContext, insList []*models.AiInferDeployInstance, needfilter bool) {
 	list := make([]*models.AiInferDeployInstance, len(insList))
 	copy(list, insList)
+
+	if needfilter {
 		for i := len(list) - 1; i >= 0; i-- {
 			if list[i].Status == constants.Running || list[i].Status == constants.Stopped {
 				list = append(list[:i], list[i+1:]...)
 			}
 		}
+	}
 
 	if len(list) == 0 {
 		return
@@ -119,3 +123,72 @@ func UpdateDeployInstanceStatus(svc *svc.ServiceContext, instance *models.AiInfe
 		return
 	}
 }
+
+func UpdateAutoStoppedInstance(svc *svc.ServiceContext) {
+	list, err := svc.Scheduler.AiStorages.GetInferDeployInstanceList()
+	if err != nil {
+		return
+	}
+
+	if len(list) == 0 {
+		return
+	}
+
+	UpdateDeployInstanceStatusBatch(svc, list, false)
+}
+
+func CheckStopStatus(in *inference.DeployInstance) bool {
+	switch in.ClusterType {
+	case storeLink.TYPE_OCTOPUS:
+		switch in.Status {
+		case "stopped":
+			return true
+		default:
+			return false
+		}
+	case storeLink.TYPE_MODELARTS:
+		switch in.Status {
+		case "stopped":
+			return true
+		default:
+			return false
+		}
+	case storeLink.TYPE_SHUGUANGAI:
+		switch in.Status {
+		case "Terminated":
+			return true
+		default:
+			return false
+		}
+	default:
+		return false
+	}
+}
+
+func CheckRunningStatus(in *inference.DeployInstance) bool {
+	switch in.ClusterType {
+	case storeLink.TYPE_OCTOPUS:
+		switch in.Status {
+		case "running":
+			return true
+		default:
+			return false
+		}
+	case storeLink.TYPE_MODELARTS:
+		switch in.Status {
+		case "running":
+			return true
+		default:
+			return false
+		}
+	case storeLink.TYPE_SHUGUANGAI:
+		switch in.Status {
+		case "Running":
+			return true
+		default:
+			return false
+		}
+	default:
+		return false
+	}
+}

internal/scheduler/service/utils/status/taskStatusSync.go

@@ -1,4 +1,4 @@
-package updater
+package status
 
 import (
 	"errors"