aiforge/modules/modelarts/modelarts.go

package modelarts

import (
	"encoding/json"
	"path"
	"strconv"

	"code.gitea.io/gitea/models"
	"code.gitea.io/gitea/modules/context"
	"code.gitea.io/gitea/modules/log"
	"code.gitea.io/gitea/modules/setting"
	"code.gitea.io/gitea/modules/storage"
)

const (
	//notebook
	storageTypeOBS   = "obs"
	autoStopDuration = 4 * 60 * 60

	DataSetMountPath = "/home/ma-user/work"
	NotebookEnv      = "Python3"
	NotebookType     = "Ascend"
	FlavorInfo       = "Ascend: 1*Ascend 910 CPU: 24 核 96GiB （modelarts.kat1.xlarge）"

	//train-job
	// ResourcePools  = "{\"resource_pool\":[{\"id\":\"pool1328035d\", \"value\":\"专属资源池\"}]}"
	// Engines        = "{\"engine\":[{\"id\":1, \"value\":\"Ascend-Powered-Engine\"}]}"
	// EngineVersions = "{\"version\":[{\"id\":118,\"value\":\"MindSpore-1.0.0-c75-python3.7-euleros2.8-aarch64\"}," +
	// 	"{\"id\":119,\"value\":\"MindSpore-1.1.1-c76-python3.7-euleros2.8-aarch64\"}," +
	// 	"{\"id\":120,\"value\":\"MindSpore-1.1.1-c76-tr5-python3.7-euleros2.8-aarch64\"}," +
	// 	"{\"id\":117,\"value\":\"TF-1.15-c75-python3.7-euleros2.8-aarch64\"}" +
	// 	"]}"
	// TrainJobFlavorInfo = "{\"flavor\":[{\"code\":\"modelarts.bm.910.arm.public.2\",\"value\":\"Ascend : 2 * Ascend 910 CPU：48 核 512GiB\"}," +
	// 	"{\"code\":\"modelarts.bm.910.arm.public.8\",\"value\":\"Ascend : 8 * Ascend 910 CPU：192 核 2048GiB\"}," +
	// 	"{\"code\":\"modelarts.bm.910.arm.public.4\",\"value\":\"Ascend : 4 * Ascend 910 CPU：96 核 1024GiB\"}," +
	// 	"{\"code\":\"modelarts.bm.910.arm.public.1\",\"value\":\"Ascend : 1 * Ascend 910 CPU：24 核 256GiB\"}" +
	// 	"]}"
	CodePath   = "/code/"
	OutputPath = "/output/"
	LogPath    = "/log/"
	JobPath    = "/job/"
	OrderDesc  = "desc" //向下查询
	OrderAsc   = "asc"  //向上查询
	Lines      = 20
	TrainUrl   = "train_url"
	DataUrl    = "data_url"
	PerPage    = 10

	SortByCreateTime = "create_time"
	ConfigTypeCustom = "custom"
)

var (
	poolInfos   *models.PoolInfos
	FlavorInfos *models.FlavorInfos
)

type GenerateTrainJobReq struct {
	JobName          string
	Uuid             string
	Description      string
	CodeObsPath      string
	BootFile         string
	DataUrl          string
	TrainUrl         string
	FlavorCode       string
	LogUrl           string
	PoolID           string
	WorkServerNumber int
	EngineID         int64
	Parameters       []models.Parameter
}

type VersionInfo struct {
	Version []struct {
		ID    int    `json:"id"`
		Value string `json:"value"`
	} `json:"version"`
}

type Flavor struct {
	Info []struct {
		Code  string `json:"code"`
		Value string `json:"value"`
	} `json:"flavor"`
}

type Engine struct {
	Info []struct {
		ID    int    `json:"id"`
		Value string `json:"value"`
	} `json:"engine"`
}

type ResourcePool struct {
	Info []struct {
		ID    string `json:"id"`
		Value string `json:"value"`
	} `json:"resource_pool"`
}

func GenerateTask(ctx *context.Context, jobName, uuid, description string) error {
	var dataActualPath string
	if uuid != "" {
		dataActualPath = setting.Bucket + "/" + setting.BasePath + path.Join(uuid[0:1], uuid[1:2]) + "/" + uuid + "/"
	} else {
		userPath := setting.UserBasePath + ctx.User.Name + "/"
		isExist, err := storage.ObsHasObject(userPath)
		if err != nil {
			log.Error("ObsHasObject failed:%v", err.Error(), ctx.Data["MsgID"])
			return err
		}

		if !isExist {
			if err = storage.ObsCreateObject(userPath); err != nil {
				log.Error("ObsCreateObject failed:%v", err.Error(), ctx.Data["MsgID"])
				return err
			}
		}

		dataActualPath = setting.Bucket + "/" + userPath
	}

	if poolInfos == nil {
		json.Unmarshal([]byte(setting.PoolInfos), &poolInfos)
	}
	jobResult, err := CreateJob(models.CreateNotebookParams{
		JobName:     jobName,
		Description: description,
		ProfileID:   setting.ProfileID,
		Flavor:      setting.Flavor,
		Pool: models.Pool{
			ID:   poolInfos.PoolInfo[0].PoolId,
			Name: poolInfos.PoolInfo[0].PoolName,
			Type: poolInfos.PoolInfo[0].PoolType,
		},
		Spec: models.Spec{
			Storage: models.Storage{
				Type: storageTypeOBS,
				Location: models.Location{
					Path: dataActualPath,
				},
			},
			AutoStop: models.AutoStop{
				Enable:   true,
				Duration: autoStopDuration,
			},
		},
	})
	if err != nil {
		log.Error("CreateJob failed: %v", err.Error())
		return err
	}
	err = models.CreateCloudbrain(&models.Cloudbrain{

		Status:  string(models.JobWaiting),
		UserID:  ctx.User.ID,
		RepoID:  ctx.Repo.Repository.ID,
		JobID:   jobResult.ID,
		JobName: jobName,
		JobType: string(models.JobTypeDebug),
		Type:    models.TypeCloudBrainNotebook,
		Uuid:    uuid,
	})

	if err != nil {
		return err
	}

	return nil
}

func GenerateTrainJob(ctx *context.Context, req *GenerateTrainJobReq) error {
	jobResult, err := createTrainJob(models.CreateTrainJobParams{
		JobName:     req.JobName,
		Description: req.Description,
		Config: models.Config{
			WorkServerNum: req.WorkServerNumber,
			AppUrl:        req.CodeObsPath,
			BootFileUrl:   req.BootFile,
			DataUrl:       req.DataUrl,
			EngineID:      req.EngineID,
			TrainUrl:      req.TrainUrl,
			LogUrl:        req.LogUrl,
			PoolID:        req.PoolID,
			CreateVersion: true,
			Flavor: models.Flavor{
				Code: req.FlavorCode,
			},
			Parameter: req.Parameters,
		},
	})
	if err != nil {
		log.Error("CreateJob failed: %v", err.Error())
		return err
	}

	attach, err := models.GetAttachmentByUUID(req.Uuid)
	if err != nil {
		log.Error("GetAttachmentByUUID(%s) failed:%v", strconv.FormatInt(jobResult.JobID, 10), err.Error())
		return nil
	}

	err = models.CreateCloudbrain(&models.Cloudbrain{
		Status:      TransTrainJobStatus(jobResult.Status),
		UserID:      ctx.User.ID,
		RepoID:      ctx.Repo.Repository.ID,
		JobID:       strconv.FormatInt(jobResult.JobID, 10),
		JobName:     req.JobName,
		JobType:     string(models.JobTypeDebug),
		Type:        models.TypeCloudBrainTrainJob,
		VersionID:   jobResult.VersionID,
		VersionName: jobResult.VersionName,
		Uuid:        req.Uuid,
		DatasetName: attach.Name,
	})

	if err != nil {
		log.Error("CreateCloudbrain(%s) failed:%v", req.JobName, err.Error())
		return err
	}

	return nil
}

func TransTrainJobStatus(status int) string {
	switch status {
	case 0:
		return "UNKNOWN"
	case 1:
		return "INIT"
	case 2:
		return "IMAGE_CREATING"
	case 3:
		return "IMAGE_FAILED"
	case 4:
		return "SUBMIT_TRYING"
	case 5:
		return "SUBMIT_FAILED"
	case 6:
		return "DELETE_FAILED"
	case 7:
		return "WAITING"
	case 8:
		return "RUNNING"
	case 9:
		return "KILLING"
	case 10:
		return "COMPLETED"
	case 11:
		return "FAILED"
	case 12:
		return "KILLED"
	case 13:
		return "CANCELED"
	case 14:
		return "LOST"
	case 15:
		return "SCALING"
	case 16:
		return "SUBMIT_MODEL_FAILED"
	case 17:
		return "DEPLOY_SERVICE_FAILED"
	case 18:
		return "CHECK_INIT"
	case 19:
		return "CHECK_RUNNING"
	case 20:
		return "CHECK_RUNNING_COMPLETED"
	case 21:
		return "CHECK_FAILED"

	default:
		return strconv.Itoa(status)
	}

	return ""
}