updated shuguang imageinfer func

Former-commit-id: 5383e3c264

desc/core/pcm-core.api

@@ -858,7 +858,7 @@ type (
 		Token         string `json:"token,optional"`
 		Ak            string `json:"ak,optional"`
 		Sk            string `json:"sk,optional"`
-		Region        string `json:"region,optional"`
+		Region        []string `json:"region,optional"`
 		ProjectId     string `json:"projectId,optional"`
 		Version       string `json:"version,optional"`
 		Label         string `json:"label,optional"`
@@ -866,6 +866,7 @@ type (
 		AuthType      string `json:"authType,optional"`
 		ProducerDict  string `json:"producerDict,optional"`
 		RegionDict    string `json:"regionDict,optional"`
+		RegionName    string `json:"regionName,optional"`
 	}
 	ClusterInfo {
 		Id            string `json:"id,omitempty" db:"id"`
@@ -888,6 +889,7 @@ type (
 		AuthType      string `json:"authType,omitempty" db:"auth_type"`
 		ProducerDict  string `json:"producerDict,omitempty" db:"producer_dict"`
 		RegionDict    string `json:"regionDict,omitempty" db:"region_dict"`
+		Location    string `json:"location,omitempty" db:"location"`
 		CreateTime    string `json:"createTime,omitempty" db:"created_time" gorm:"autoCreateTime"`
 	}
 )

go.mod

@@ -18,8 +18,8 @@ require (
 	github.com/prometheus/common v0.59.1
 	github.com/robfig/cron/v3 v3.0.1
 	github.com/zeromicro/go-zero v1.7.2
-	gitlink.org.cn/JointCloud/pcm-ac v0.0.0-20240918015229-59c579d1a437
-	gitlink.org.cn/JointCloud/pcm-modelarts v0.0.0-20240909072501-939c3144cd9e
+	gitlink.org.cn/JointCloud/pcm-ac v0.0.0-20240920093406-601f283f0185
+	gitlink.org.cn/JointCloud/pcm-modelarts v0.0.0-20240918011543-482dcd609877
 	gitlink.org.cn/JointCloud/pcm-octopus v0.0.0-20240817071412-44397870b110
 	gitlink.org.cn/JointCloud/pcm-openstack v0.0.0-20240403033338-e7edabad4203
 	gitlink.org.cn/JointCloud/pcm-slurm v0.0.0-20240301080743-8b94bbaf57f5

go.sum

@@ -466,10 +466,10 @@ github.com/yuin/gopher-lua v1.1.1 h1:kYKnWBjvbNP4XLT3+bPEwAXJx262OhaHDWDVOPjL46M
 github.com/yuin/gopher-lua v1.1.1/go.mod h1:GBR0iDaNXjAgGg9zfCvksxSRnQx76gclCIb7kdAd1Pw=
 github.com/zeromicro/go-zero v1.7.2 h1:a8lyVOG3KXG4LrAy6ZmtJTJtisX4Ostc4Pst4fE704I=
 github.com/zeromicro/go-zero v1.7.2/go.mod h1:WFXfF92Exw0O7WECifS6r99JSzv4KEN49x9RhAfgkMc=
-gitlink.org.cn/JointCloud/pcm-ac v0.0.0-20240918015229-59c579d1a437 h1:ta6h9+FU7AQ2fNyQiXrZnMdlNBjOKdyBx4e3RF7BE84=
-gitlink.org.cn/JointCloud/pcm-ac v0.0.0-20240918015229-59c579d1a437/go.mod h1:3eECiw9O2bIFkkePlloKyLNXiqBAhOxNrDoGaaGseGY=
-gitlink.org.cn/JointCloud/pcm-modelarts v0.0.0-20240909072501-939c3144cd9e h1:6LYJggBoeAQxy/otzWjt40Pa7gnVvUR4c5YMi6A/NdU=
-gitlink.org.cn/JointCloud/pcm-modelarts v0.0.0-20240909072501-939c3144cd9e/go.mod h1:/eOmBFZKWGoabG3sRVkVvIbLwsd2631k4jkUBR6x1AA=
+gitlink.org.cn/JointCloud/pcm-ac v0.0.0-20240920093406-601f283f0185 h1:B+YBB5xHlIAS6ILuaCGQwbOpr/L6LOHAlj9PeFUCetM=
+gitlink.org.cn/JointCloud/pcm-ac v0.0.0-20240920093406-601f283f0185/go.mod h1:3eECiw9O2bIFkkePlloKyLNXiqBAhOxNrDoGaaGseGY=
+gitlink.org.cn/JointCloud/pcm-modelarts v0.0.0-20240918011543-482dcd609877 h1:a+1FpxqLPRojlAkJlAeRhKRbxajymXYgrM+s9bfQx0E=
+gitlink.org.cn/JointCloud/pcm-modelarts v0.0.0-20240918011543-482dcd609877/go.mod h1:/eOmBFZKWGoabG3sRVkVvIbLwsd2631k4jkUBR6x1AA=
 gitlink.org.cn/JointCloud/pcm-octopus v0.0.0-20240817071412-44397870b110 h1:GaXwr5sgDh0raHjUf9IewTvnRvajYea7zbLsaerYyXo=
 gitlink.org.cn/JointCloud/pcm-octopus v0.0.0-20240817071412-44397870b110/go.mod h1:QOD5+/l2D+AYBjF2h5T0mdJyfGAmF78QmeKdbBXbjLQ=
 gitlink.org.cn/JointCloud/pcm-openstack v0.0.0-20240403033338-e7edabad4203 h1:s6PsZ1+bev294IWdZRlV7mnOwI1+UzFcldVW/BqhQzI=

internal/logic/adapters/createclusterlogic.go

@@ -3,12 +3,16 @@ package adapters
 import (
 	"context"
 	"errors"
-	"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/utils"
-	"gorm.io/gorm"
-	"time"
-
+	"fmt"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
+	tool "gitlink.org.cn/JointCloud/pcm-coordinator/pkg/utils"
+	"gorm.io/gorm"
+	"io/ioutil"
+	"k8s.io/apimachinery/pkg/util/json"
+	"net/http"
+	"net/url"
+	"time"
 
 	"github.com/zeromicro/go-zero/core/logx"
 )
@@ -33,17 +37,27 @@ func (l *CreateClusterLogic) CreateCluster(req *types.ClusterCreateReq) (resp *t
 	if errors.Is(result.Error, gorm.ErrRecordNotFound) {
 		return nil, errors.New("adapter does not exist")
 	}
+
 	cluster := types.ClusterInfo{}
-	utils.Convert(req, &cluster)
-	cluster.Id = utils.GenSnowflakeIDStr()
+
+	tool.Convert(req, &cluster)
 	cluster.CreateTime = time.Now().Format("2006-01-02 15:04:05")
 	cluster.OwnerId = "0"
+	// 获取集群经纬度
+	location, err := GeoMap(req.RegionName)
+	cluster.Location = location
+
+	cluster.Id = tool.GenSnowflakeIDStr()
 	tx := l.svcCtx.DbEngin.Table("t_cluster").Create(&cluster)
 	if tx.Error != nil {
 		logx.Errorf(tx.Error.Error())
 		return nil, errors.New("cluster create failed")
 	}
 
+	if err != nil {
+		return nil, err
+	}
+
 	// push cluster info to adapter
 	var adapterServer string
 	l.svcCtx.DbEngin.Raw("select server from t_adapter  where id = ?", req.AdapterId).Scan(&adapterServer)
@@ -64,3 +78,63 @@ func (l *CreateClusterLogic) CreateCluster(req *types.ClusterCreateReq) (resp *t
 	}
 	return
 }
+
+func GeoMap(address string) (string, error) {
+	// 此处填写您在控制台-应用管理-创建应用后获取的AK
+	ak := "d3cc9eee0266d39a52498726d1b82f87"
+
+	// 接口地址
+	uri := "https://restapi.amap.com/v3/geocode/geo"
+
+	// 设置请求参数
+	params := url.Values{
+		"address": []string{address},
+		"output":  []string{"json"},
+		"key":     []string{ak},
+	}
+
+	// 发起请求
+	request, err := url.Parse(uri + "?" + params.Encode())
+	if nil != err {
+		fmt.Printf("host error: %v", err)
+		return "", err
+	}
+
+	resp, err1 := http.Get(request.String())
+	fmt.Printf("url: %s\n", request.String())
+	defer resp.Body.Close()
+	if err1 != nil {
+		fmt.Printf("request error: %v", err1)
+		return "", err
+	}
+	body, err2 := ioutil.ReadAll(resp.Body)
+	if err2 != nil {
+		fmt.Printf("response error: %v", err2)
+	}
+
+	fmt.Println(string(body))
+	geoResponse := GeoResponse{}
+	json.Unmarshal(body, &geoResponse)
+	return geoResponse.Geocodes[0].Location, err
+}
+
+type GeoResponse struct {
+	Status   string    `json:"status"`
+	Info     string    `json:"info"`
+	InfoCode string    `json:"infocode"`
+	Count    string    `json:"count"`
+	Geocodes []GeoCode `json:"geocodes"`
+}
+
+type GeoCode struct {
+	FormattedAddress string `json:"formatted_address"`
+	Country          string `json:"country"`
+	Province         string `json:"province"`
+	CityCode         string `json:"citycode"`
+	City             string `json:"city"`
+	District         string `json:"district"`
+	Adcode           string `json:"adcode"`
+	Number           string `json:"number"`
+	Location         string `json:"location"`
+	Level            string `json:"level"`
+}

internal/scheduler/service/inference/imageInference/imageInference.go

@@ -463,7 +463,7 @@ func getInferResult(url string, file multipart.File, fileName string, clusterId
 	switch clusterType {
 	case storeLink.TYPE_OCTOPUS:
 		r := http.Request{}
-		result, err := iCluster.GetInferResult(r.Context(), url, file, fileName)
+		result, err := iCluster.GetImageInferResult(r.Context(), url, file, fileName)
 		if err != nil {
 			return "", err
 		}

internal/scheduler/service/inference/inference.go

@@ -12,13 +12,17 @@ const (
 
 type ICluster interface {
 	GetClusterInferUrl(ctx context.Context, option *option.InferOption) (*ClusterInferUrl, error)
-	GetInferResult(ctx context.Context, url string, file multipart.File, fileName string) (string, error)
 	GetInferDeployInstanceList(ctx context.Context) ([]*DeployInstance, error)
 	StartInferDeployInstance(ctx context.Context, id string) bool
 	StopInferDeployInstance(ctx context.Context, id string) bool
 	GetInferDeployInstance(ctx context.Context, id string) (*DeployInstance, error)
 	CreateInferDeployInstance(ctx context.Context, option *option.InferOption) (string, error)
 	CheckModelExistence(ctx context.Context, modelName string, modelType string) bool
+	InferResult
+}
+
+type InferResult interface {
+	GetImageInferResult(ctx context.Context, url string, file multipart.File, fileName string) (string, error)
 }
 
 type IInference interface {

internal/storeLink/modelarts.go

@@ -767,7 +767,7 @@ func (m *ModelArtsLink) GetInferDeployInstance(ctx context.Context, id string) (
 	return ins, nil
 }
 
-func (m *ModelArtsLink) GetInferResult(ctx context.Context, url string, file multipart.File, fileName string) (string, error) {
+func (m *ModelArtsLink) GetImageInferResult(ctx context.Context, url string, file multipart.File, fileName string) (string, error) {
 	return "", nil
 }
 
@@ -791,11 +791,14 @@ func (m *ModelArtsLink) CreateInferDeployInstance(ctx context.Context, option *o
 	}
 	var configItems []*modelarts.ServiceConfig
 	configItems = append(configItems, configParam)
+	now := time.Now()
+	timestampSec := now.Unix()
+	str := strconv.FormatInt(timestampSec, 10)
 	req := &modelarts.CreateServiceReq{
 		Platform:    m.platform,
 		Config:      configItems,
 		InferType:   "real-time",
-		ServiceName: option.ModelName + "_" + option.ModelType + "_" + Npu,
+		ServiceName: option.ModelName + "_" + option.ModelType + "_" + Npu + "_" + str,
 	}
 	ctx, cancel := context.WithTimeout(context.Background(), 150*time.Second)
 	defer cancel()

internal/storeLink/octopus.go

@@ -1183,7 +1183,7 @@ func (o *OctopusLink) GetInferDeployInstance(ctx context.Context, id string) (*i
 	return ins, nil
 }
 
-func (o *OctopusLink) GetInferResult(ctx context.Context, url string, file multipart.File, fileName string) (string, error) {
+func (o *OctopusLink) GetImageInferResult(ctx context.Context, url string, file multipart.File, fileName string) (string, error) {
 	stream, err := o.octopusRpc.GetInferResult(ctx)
 	if err != nil {
 		return "", err

internal/storeLink/shuguangai.go

@@ -47,49 +47,60 @@ const (
 	PythonCodePath        = "/work/home/acgnnmfbwo/111111/py/test.py"
 	DATASETS_DIR          = "/work/home/acgnnmfbwo/pcmv1/dataset"
 	ALGORITHM_DIR         = "/work/home/acgnnmfbwo/pcmv1/algorithm"
+	KUNSHAN_DIR           = "/public/home/acgnnmfbwo/pcmv1"
 	TRAIN_FILE            = "train.py"
 	CPUCOREPRICEPERHOUR   = 0.09
 	DCUPRICEPERHOUR       = 2.0
 	KB                    = 1024
 	TIMEOUT               = 20
 	DEPLOY_INSTANCE_LIMIT = 100
+	ProtocolType          = "HTTP"
+	ContainerPort         = 8881
+	JUPYTER               = "jupyter"
 )
 
-var RESOURCESGAIMAP = map[string]ResourceSpecSGAI{
-	"WodTB2rJ8SobMgQ1nrtR245jxOrsovFi": {
-		CPU: 1,
-		GPU: 1,
-		RAM: 2 * RAM_SIZE_1G,
-	},
-	"6d41v1XV53MQPmQOJ5kNatIck9yl8nWZ": {
-		CPU: 1,
-		GPU: 2,
-		RAM: 2 * RAM_SIZE_1G,
-	},
-	"OBtVaaXAv9n9FbLR7pWAoa3yR13jXwNc": {
-		CPU: 2,
-		GPU: 3,
-		RAM: 4 * RAM_SIZE_1G,
-	},
-	"sBWfpkntUzsWYly11kdwEHZOYYIsFmve": {
-		CPU: 4,
-		GPU: 4,
-		RAM: 8 * RAM_SIZE_1G,
-	},
-	"jeYBVPwyIALjVYNzHvysh2o5CsBpBLp2": {
-		CPU: 5,
-		GPU: 5,
-		RAM: 10 * RAM_SIZE_1G,
-	},
-}
+var (
+	RESOURCESGAIMAP = map[string]ResourceSpecSGAI{
+		"WodTB2rJ8SobMgQ1nrtR245jxOrsovFi": {
+			CPU: 1,
+			GPU: 1,
+			RAM: 2 * RAM_SIZE_1G,
+		},
+		"6d41v1XV53MQPmQOJ5kNatIck9yl8nWZ": {
+			CPU: 1,
+			GPU: 2,
+			RAM: 2 * RAM_SIZE_1G,
+		},
+		"OBtVaaXAv9n9FbLR7pWAoa3yR13jXwNc": {
+			CPU: 2,
+			GPU: 3,
+			RAM: 4 * RAM_SIZE_1G,
+		},
+		"sBWfpkntUzsWYly11kdwEHZOYYIsFmve": {
+			CPU: 4,
+			GPU: 4,
+			RAM: 8 * RAM_SIZE_1G,
+		},
+		"jeYBVPwyIALjVYNzHvysh2o5CsBpBLp2": {
+			CPU: 5,
+			GPU: 5,
+			RAM: 10 * RAM_SIZE_1G,
+		},
+	}
 
-var RESOURCESPECSAI = map[string]string{
-	"WodTB2rJ8SobMgQ1nrtR245jxOrsovFi": "CPU:1, DCU:1, RAM:2G",
-	"6d41v1XV53MQPmQOJ5kNatIck9yl8nWZ": "CPU:1, DCU:2, RAM:2G",
-	"OBtVaaXAv9n9FbLR7pWAoa3yR13jXwNc": "CPU:2, DCU:3, RAM:4G",
-	"sBWfpkntUzsWYly11kdwEHZOYYIsFmve": "CPU:4, DCU:4, RAM:8G",
-	"jeYBVPwyIALjVYNzHvysh2o5CsBpBLp2": "CPU:5, DCU:5, RAM:10G",
-}
+	RESOURCESPECSAI = map[string]string{
+		"WodTB2rJ8SobMgQ1nrtR245jxOrsovFi": "CPU:1, DCU:1, RAM:2G",
+		"6d41v1XV53MQPmQOJ5kNatIck9yl8nWZ": "CPU:1, DCU:2, RAM:2G",
+		"OBtVaaXAv9n9FbLR7pWAoa3yR13jXwNc": "CPU:2, DCU:3, RAM:4G",
+		"sBWfpkntUzsWYly11kdwEHZOYYIsFmve": "CPU:4, DCU:4, RAM:8G",
+		"jeYBVPwyIALjVYNzHvysh2o5CsBpBLp2": "CPU:5, DCU:5, RAM:10G",
+	}
+
+	ModelNameCmdMap = map[string]string{
+		"blip-image-captioning-base": "pip install transformers python-multipart fastapi uvicorn[standard]; python /public/home/acgnnmfbwo/pcmv1/inference/pytorch/blip_image_captioning_base/infer.py",
+		"imagenet_resnet50":          "pip install fastapi uvicorn[standard] python-multipart; python /public/home/acgnnmfbwo/pcmv1/inference/pytorch/imagenet_resnet50/infer.py",
+	}
+)
 
 type ResourceSpecSGAI struct {
 	CPU int64
@@ -905,15 +916,81 @@ func (s *ShuguangAi) GetInferDeployInstance(ctx context.Context, id string) (*in
 	return ins, nil
 }
 
-func (s *ShuguangAi) GetInferResult(ctx context.Context, url string, file multipart.File, fileName string) (string, error) {
+func (s *ShuguangAi) GetImageInferResult(ctx context.Context, url string, file multipart.File, fileName string) (string, error) {
 	return "", nil
 }
 
 func (s *ShuguangAi) CreateInferDeployInstance(ctx context.Context, option *option.InferOption) (string, error) {
+	containerPortInfoList := []*hpcAC.ContainerPortInfoList{
+		{
+			ProtocolType:  ProtocolType,
+			ContainerPort: ContainerPort,
+		},
+	}
 
-	return "", nil
+	desc := option.ModelType + FORWARD_SLASH + option.ModelName + FORWARD_SLASH + strings.ToLower(DCU)
+	instanceServiceName := "infer_instance" + UNDERSCORE + utils.RandomString(15)
+	resourceGroup := "kshdtest"
+
+	script, ok := ModelNameCmdMap[option.ModelName]
+	if !ok {
+		return "", errors.New("failed to set cmd, ModelName not exist")
+	}
+
+	param := &hpcAC.CreateParams{
+		AcceleratorType:       strings.ToLower(DCU),
+		ContainerPortInfoList: containerPortInfoList,
+		CpuNumber:             8,
+		Description:           desc,
+		//env
+		GpuNumber:           1,
+		ImagePath:           "11.11.100.6:5000/dcu/admin/base/jupyterlab-pytorch:1.13.1-py3.7-dtk23.04-centos7.6",
+		InstanceServiceName: instanceServiceName,
+		MountInfoList:       make([]*hpcAC.MountInfoList, 0),
+		//originalVersion
+		RamSize: 10 * RAM_SIZE_1G,
+		//rdma
+		ResourceGroup:          resourceGroup,
+		StartScriptActionScope: "all",
+		StartScriptContent:     script,
+		//startServiceCommand
+		//taskClassification: "interactive"
+		TaskNumber:     1,
+		TaskType:       JUPYTER,
+		TimeoutLimit:   "01:00:00",
+		UseStartScript: true,
+		//useStartServiceCommand: false
+		Version: "jupyterlab-pytorch:1.13.1-py3.7-dtk23.04-centos7.6",
+	}
+
+	req := &hpcacclient.CreateInstanceServiceReq{
+		Data: param,
+	}
+
+	resp, err := s.aCRpc.CreateInstanceService(ctx, req)
+	if err != nil {
+		return "", err
+	}
+	if resp.Code != "0" {
+		return "", errors.New(resp.Msg)
+	}
+
+	return resp.Data, nil
 }
 
 func (s *ShuguangAi) CheckModelExistence(ctx context.Context, name string, mtype string) bool {
-	return false
+	modelPath := "model" + FORWARD_SLASH + name
+	req := &hpcAC.IsExistFileReq{
+		Path: KUNSHAN_DIR + FORWARD_SLASH + modelPath,
+	}
+	resp, err := s.aCRpc.IsExistFile(ctx, req)
+	if err != nil {
+		return false
+	}
+
+	if resp.Code != "0" || resp.Data == nil {
+		return false
+	}
+
+	return resp.Data.Exist
 }

internal/types/types.go

@@ -779,26 +779,27 @@ type ClusterReq struct {
 }
 
 type ClusterCreateReq struct {
-	Id            string `json:"id,optional"`
-	AdapterId     string `json:"adapterId,optional"`
-	Name          string `json:"name,optional"`
-	Nickname      string `json:"nickname,optional"`
-	Description   string `json:"description,optional"`
-	Server        string `json:"server,optional"`
-	MonitorServer string `json:"monitorServer,optional"`
-	Username      string `json:"username,optional"`
-	Password      string `json:"password,optional"`
-	Token         string `json:"token,optional"`
-	Ak            string `json:"ak,optional"`
-	Sk            string `json:"sk,optional"`
-	Region        string `json:"region,optional"`
-	ProjectId     string `json:"projectId,optional"`
-	Version       string `json:"version,optional"`
-	Label         string `json:"label,optional"`
-	OwnerId       string `json:"ownerId,omitempty,optional"`
-	AuthType      string `json:"authType,optional"`
-	ProducerDict  string `json:"producerDict,optional"`
-	RegionDict    string `json:"regionDict,optional"`
+	Id            string   `json:"id,optional"`
+	AdapterId     string   `json:"adapterId,optional"`
+	Name          string   `json:"name,optional"`
+	Nickname      string   `json:"nickname,optional"`
+	Description   string   `json:"description,optional"`
+	Server        string   `json:"server,optional"`
+	MonitorServer string   `json:"monitorServer,optional"`
+	Username      string   `json:"username,optional"`
+	Password      string   `json:"password,optional"`
+	Token         string   `json:"token,optional"`
+	Ak            string   `json:"ak,optional"`
+	Sk            string   `json:"sk,optional"`
+	Region        []string `json:"region,optional"`
+	ProjectId     string   `json:"projectId,optional"`
+	Version       string   `json:"version,optional"`
+	Label         string   `json:"label,optional"`
+	OwnerId       string   `json:"ownerId,omitempty,optional"`
+	AuthType      string   `json:"authType,optional"`
+	ProducerDict  string   `json:"producerDict,optional"`
+	RegionDict    string   `json:"regionDict,optional"`
+	RegionName    string   `json:"regionName,optional"`
 }
 
 type ClusterInfo struct {
@@ -822,6 +823,7 @@ type ClusterInfo struct {
 	AuthType      string `json:"authType,omitempty" db:"auth_type"`
 	ProducerDict  string `json:"producerDict,omitempty" db:"producer_dict"`
 	RegionDict    string `json:"regionDict,omitempty" db:"region_dict"`
+	Location      string `json:"location,omitempty" db:"location"`
 	CreateTime    string `json:"createTime,omitempty" db:"created_time" gorm:"autoCreateTime"`
 }