Merge remote-tracking branch 'origin/master'

Former-commit-id: c797cc12a5

internal/logic/inference/deployinstancelistlogic.go

@@ -4,9 +4,11 @@ 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/svc"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/models"
+	"time"
 )
 
 type DeployInstanceListLogic struct {
@@ -36,6 +38,17 @@ func (l *DeployInstanceListLogic) DeployInstanceList(req *types.DeployInstanceLi
 		return nil, tx.Error
 	}
 
+	ins := list[0]
+	for i := range list {
+		last, _ := time.Parse(time.RFC3339, ins.UpdateTime)
+		latest, _ := time.Parse(time.RFC3339, list[i].UpdateTime)
+		if latest.After(last) {
+			ins = list[i]
+		}
+	}
+
+	go updater.UpdateDeployInstanceStatus(l.svcCtx, ins)
+
 	//count total
 	var total int64
 	err = tx.Count(&total).Error

internal/logic/inference/startdeployinstancelistlogic.go

@@ -26,6 +26,10 @@ func NewStartDeployInstanceListLogic(ctx context.Context, svcCtx *svc.ServiceCon
 func (l *StartDeployInstanceListLogic) StartDeployInstanceList(req *types.StartDeployInstanceReq) (resp *types.StartDeployInstanceResp, err error) {
 	resp = &types.StartDeployInstanceResp{}
 	success := l.svcCtx.Scheduler.AiService.InferenceAdapterMap[req.AdapterId][req.ClusterId].StartInferDeployInstance(l.ctx, req.InstanceId)
+	if !success {
+		return nil, errors.New("start instance failed")
+	}
+
 	id, err := strconv.ParseInt(req.Id, 10, 64)
 	ins, err := l.svcCtx.Scheduler.AiStorages.GetInferDeployInstanceById(id)
 	if err != nil {
@@ -34,9 +38,5 @@ func (l *StartDeployInstanceListLogic) StartDeployInstanceList(req *types.StartD
 
 	l.svcCtx.Scheduler.AiStorages.UpdateInferDeployInstance(ins)
 
-	if !success {
-		return nil, errors.New("start instance failed")
-	}
-
 	return resp, nil
 }

internal/logic/inference/stopdeployinstancelogic.go

@@ -26,6 +26,10 @@ func NewStopDeployInstanceLogic(ctx context.Context, svcCtx *svc.ServiceContext)
 func (l *StopDeployInstanceLogic) StopDeployInstance(req *types.StopDeployInstanceReq) (resp *types.StopDeployInstanceResp, err error) {
 	resp = &types.StopDeployInstanceResp{}
 	success := l.svcCtx.Scheduler.AiService.InferenceAdapterMap[req.AdapterId][req.ClusterId].StopInferDeployInstance(l.ctx, req.InstanceId)
+	if !success {
+		return nil, errors.New("stop instance failed")
+	}
+
 	id, err := strconv.ParseInt(req.Id, 10, 64)
 	ins, err := l.svcCtx.Scheduler.AiStorages.GetInferDeployInstanceById(id)
 	if err != nil {
@@ -34,8 +38,5 @@ func (l *StopDeployInstanceLogic) StopDeployInstance(req *types.StopDeployInstan
 
 	l.svcCtx.Scheduler.AiStorages.UpdateInferDeployInstance(ins)
 
-	if !success {
-		return nil, errors.New("stop instance failed")
-	}
 	return resp, nil
 }

internal/scheduler/database/aiStorage.go

@@ -481,7 +481,7 @@ func (s *AiStorage) GetTrainingTaskTotalNum() (int32, error) {
 
 func (s *AiStorage) GetTrainingTaskRunningNum() (int32, error) {
 	var total int32
-	tx := s.DbEngin.Raw("select count(*) from task where `task_type_dict` = 11 and `status` = 'Running'").Scan(&total)
+	tx := s.DbEngin.Raw("select count(*) from task where `task_type_dict` = 10 and `status` = 'Running'").Scan(&total)
 	if tx.Error != nil {
 		logx.Errorf(tx.Error.Error())
 		return 0, tx.Error