存算联动添加曙光Hpc

api/internal/storeLink/shuguangHpc.go

@@ -0,0 +1,48 @@
+package storeLink
+
+import (
+	"context"
+	"gitlink.org.cn/jcce-pcm/pcm-coordinator/api/internal/svc"
+	"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/models"
+)
+
+type ShuguangHpc struct {
+	ctx         context.Context
+	svcCtx      *svc.ServiceContext
+	participant *models.StorelinkCenter
+}
+
+func (s ShuguangHpc) UploadImage(path string) (interface{}, error) {
+	//TODO implement me
+	panic("implement me")
+}
+
+func (s ShuguangHpc) DeleteImage(imageId string) (interface{}, error) {
+	//TODO implement me
+	panic("implement me")
+}
+
+func (s ShuguangHpc) QueryImageList() (interface{}, error) {
+	//TODO implement me
+	panic("implement me")
+}
+
+func (s ShuguangHpc) SubmitTask(imageId string, cmd string, envs []string, params []string, resourceId string) (interface{}, error) {
+	//TODO implement me
+	panic("implement me")
+}
+
+func (s ShuguangHpc) QueryTask(taskId string) (interface{}, error) {
+	//TODO implement me
+	panic("implement me")
+}
+
+func (s ShuguangHpc) QuerySpecs() (interface{}, error) {
+	//TODO implement me
+	panic("implement me")
+}
+
+func (s ShuguangHpc) DeleteTask(taskId string) (interface{}, error) {
+	//TODO implement me
+	panic("implement me")
+}

api/internal/storeLink/shuguangai.go

@@ -37,6 +37,14 @@ const (
 	WORKER_GPU_NUMBER               = 1
 	SHUGUANGAI_CUSTOM_RESOURCE_ID   = "WodTB2rJ8SobMgQ1nrtR245jxOrsovFi"
 	SHUGUANGAI_CUSTOM_RESOURCE_NAME = "1*DCU, CPU:5, 内存:10GB"
+	DCU                             = "dcu"
+	PYTORCH                         = "Pytorch"
+	TASK_PYTORCH_PREFIX             = "PytorchTask"
+	TENSORFLOW                      = "Tensorflow"
+	RESOURCE_GROUP                  = "wzhdtest"
+	WorkPath                        = "/work/home/acgnnmfbwo/111111/py/"
+	TimeoutLimit                    = "10:00:00"
+	PythonCodePath                  = "/work/home/acgnnmfbwo/111111/py/test.py"
 )
 
 func NewShuguangAi(ctx context.Context, svcCtx *svc.ServiceContext, participant *models.StorelinkCenter) *ShuguangAi {

api/internal/storeLink/storeLink.go

@@ -38,24 +38,16 @@ type Linkage interface {
 }
 
 const (
-	PY_PARAM_PREFIX     = "--"
-	SPACE               = " "
-	UNDERSCORE          = "_"
-	COMMA               = ","
-	TYPE_OCTOPUS        = "1"
-	TYPE_MODELARTS      = "2"
-	TYPE_SHUGUANGAI     = "3"
-	OCTOPUS             = "Octopus"
-	MODELARTS           = "Modelarts"
-	SHUGUANGAI          = "ShuguangAi"
-	DCU                 = "dcu"
-	PYTORCH             = "Pytorch"
-	TASK_PYTORCH_PREFIX = "PytorchTask"
-	TENSORFLOW          = "Tensorflow"
-	RESOURCE_GROUP      = "wzhdtest"
-	WorkPath            = "/work/home/acgnnmfbwo/111111/py/"
-	TimeoutLimit        = "10:00:00"
-	PythonCodePath      = "/work/home/acgnnmfbwo/111111/py/test.py"
+	PY_PARAM_PREFIX = "--"
+	SPACE           = " "
+	UNDERSCORE      = "_"
+	COMMA           = ","
+	TYPE_OCTOPUS    = "1"
+	TYPE_MODELARTS  = "2"
+	TYPE_SHUGUANGAI = "3"
+	OCTOPUS         = "Octopus"
+	MODELARTS       = "Modelarts"
+	SHUGUANGAI      = "ShuguangAi"
 )
 
 var (

pkg/scheduler/scheduler.go

@@ -139,15 +139,16 @@ func (s *scheduler) SaveToDb() error {
 
 func (s *scheduler) obtainParamsforStrategy() (*algo.Task, []*algo.Provider, error) {
 	task, providerList := s.scheduleService.genTaskAndProviders(s.task, s.dbEngin)
+
+	if len(providerList) == 0 {
+		return nil, nil, errors.New("获取集群失败")
+	}
+
 	// 过滤可用集群
 	err := s.filterAvailableProviders(&providerList)
 	if err != nil {
 		return nil, nil, err
 	}
-	//可用集群为0
-	if len(providerList) == 0 {
-		return nil, nil, errors.New("未能获取可用集群")
-	}
 
 	return task, providerList, nil
 }