Merge branch 'zouap' of https://git.openi.org.cn/OpenI/aiforge into zouap

models/ai_model_manage.go

@@ -30,6 +30,7 @@ type AiModelManage struct {
 	CodeBranch    string             `xorm:"varchar(400) NULL"`
 	CodeCommitID  string             `xorm:"NULL"`
 	UserId        int64              `xorm:"NOT NULL"`
+	UserName      string             `xorm:"NULL"`
 	TrainTaskInfo string             `xorm:"text NULL"`
 	CreatedUnix   timeutil.TimeStamp `xorm:"created"`
 	UpdatedUnix   timeutil.TimeStamp `xorm:"INDEX updated"`

models/cloudbrain.go

@@ -31,6 +31,7 @@ const (
 	JobTypeBrainScore   JobType = "BRAINSCORE"
 	JobTypeTrain        JobType = "TRAIN"
 
+	//notebook
 	ModelArtsCreateQueue  ModelArtsJobStatus = "CREATE_QUEUING" //免费资源创建排队中
 	ModelArtsCreating     ModelArtsJobStatus = "CREATING"       //创建中
 	ModelArtsCreateFailed ModelArtsJobStatus = "CREATE_FAILED"  //创建失败
@@ -46,6 +47,30 @@ const (
 	ModelArtsDeleted      ModelArtsJobStatus = "DELETED"        //已删除
 	ModelArtsResizing     ModelArtsJobStatus = "RESIZING"       //规格变更中
 	ModelArtsResizFailed  ModelArtsJobStatus = "RESIZE_FAILED"  //规格变更失败
+
+	//trainjob
+	ModelArtsTrainJobUnknown               ModelArtsJobStatus = "UNKNOWN"                 //作业状态未知
+	ModelArtsTrainJobInit                  ModelArtsJobStatus = "INIT"                    //作业初始化状态
+	ModelArtsTrainJobImageCreating         ModelArtsJobStatus = "IMAGE_CREATING"          //作业镜像正在创建
+	ModelArtsTrainJobImageFailed           ModelArtsJobStatus = "IMAGE_FAILED"            //作业镜像创建失败
+	ModelArtsTrainJobSubmitTrying          ModelArtsJobStatus = "SUBMIT_TRYING"           //作业正在提交
+	ModelArtsTrainJobSubmitFailed          ModelArtsJobStatus = "SUBMIT_FAILED"           //作业提交失败
+	ModelArtsTrainJobDeleteFailed          ModelArtsJobStatus = "DELETE_FAILED"           //作业删除失败
+	ModelArtsTrainJobWaiting               ModelArtsJobStatus = "WAITING"                 //作业正在排队中
+	ModelArtsTrainJobRunning               ModelArtsJobStatus = "RUNNING"                 //作业正在运行中
+	ModelArtsTrainJobKilling               ModelArtsJobStatus = "KILLING"                 //作业正在取消
+	ModelArtsTrainJobCompleted             ModelArtsJobStatus = "COMPLETED"               //作业已经完成
+	ModelArtsTrainJobFailed                ModelArtsJobStatus = "FAILED"                  //作业运行失败
+	ModelArtsTrainJobKilled                ModelArtsJobStatus = "KILLED"                  //作业取消成功
+	ModelArtsTrainJobCanceled              ModelArtsJobStatus = "CANCELED"                //作业取消
+	ModelArtsTrainJobLost                  ModelArtsJobStatus = "LOST"                    //作业丢失
+	ModelArtsTrainJobScaling               ModelArtsJobStatus = "SCALING"                 //作业正在扩容
+	ModelArtsTrainJobSubmitModelFailed     ModelArtsJobStatus = "SUBMIT_MODEL_FAILED"     //提交模型失败
+	ModelArtsTrainJobDeployServiceFailed   ModelArtsJobStatus = "DEPLOY_SERVICE_FAILED"   //部署服务失败
+	ModelArtsTrainJobCheckInit             ModelArtsJobStatus = "CHECK_INIT"              //审核作业初始化
+	ModelArtsTrainJobCheckRunning          ModelArtsJobStatus = "CHECK_RUNNING"           //审核作业正在运行中
+	ModelArtsTrainJobCheckRunningCompleted ModelArtsJobStatus = "CHECK_RUNNING_COMPLETED" //审核作业已经完成
+	ModelArtsTrainJobCheckFailed           ModelArtsJobStatus = "CHECK_FAILED"            //审核作业失败
 )
 
 type Cloudbrain struct {
@@ -1133,3 +1158,14 @@ func CanDelJob(isSigned bool, user *User, job *CloudbrainInfo) bool {
 	}
 	return false
 }
+
+func GetCloudBrainUnStoppedJob() ([]*Cloudbrain, error) {
+	cloudbrains := make([]*Cloudbrain, 0, 10)
+	return cloudbrains, x.
+		NotIn("status",
+			JobStopped, JobSucceeded, JobFailed, ModelArtsCreateFailed, ModelArtsStartFailed, ModelArtsUnavailable, ModelArtsResizFailed, ModelArtsDeleted,
+			ModelArtsStopped, ModelArtsTrainJobCanceled, ModelArtsTrainJobCheckFailed, ModelArtsTrainJobCompleted, ModelArtsTrainJobDeleteFailed, ModelArtsTrainJobDeployServiceFailed,
+			ModelArtsTrainJobFailed, ModelArtsTrainJobImageFailed, ModelArtsTrainJobKilled, ModelArtsTrainJobLost, ModelArtsTrainJobSubmitFailed, ModelArtsTrainJobSubmitModelFailed).
+		Limit(100).
+		Find(&cloudbrains)
+}

modules/auth/auth.go

@@ -9,6 +9,9 @@ import (
 	"reflect"
 	"strings"
 
+	"code.gitea.io/gitea/modules/base"
+	"code.gitea.io/gitea/modules/setting"
+
 	"code.gitea.io/gitea/models"
 	"code.gitea.io/gitea/modules/auth/sso"
 	"code.gitea.io/gitea/modules/validation"
@@ -31,6 +34,8 @@ func SignedInUser(ctx *macaron.Context, sess session.Store) (*models.User, bool)
 		return nil, false
 	}
 
+	checkAutoLogin(ctx, sess)
+
 	// Try to sign in with each of the enabled plugins
 	for _, ssoMethod := range sso.Methods() {
 		if !ssoMethod.IsEnabled() {
@@ -46,6 +51,23 @@ func SignedInUser(ctx *macaron.Context, sess session.Store) (*models.User, bool)
 	return nil, false
 }
 
+func checkAutoLogin(ctx *macaron.Context, sess session.Store) {
+	uid := sess.Get("uid")
+	if uid == nil {
+		uname := ctx.GetCookie(setting.CookieUserName)
+
+		u, err := models.GetUserByName(uname)
+		if err == nil {
+
+			if val, ok := ctx.GetSuperSecureCookie(
+				base.EncodeMD5(u.Rands+u.Passwd), setting.CookieRememberName); ok && val == u.Name {
+				sess.Set("uid", u.ID)
+			}
+		}
+	}
+
+}
+
 // Form form binding interface
 type Form interface {
 	binding.Validator

modules/cron/tasks_basic.go

@@ -185,6 +185,17 @@ func registerHandleSummaryStatistic() {
 	})
 }
 
+func registerSyncCloudbrainStatus() {
+	RegisterTaskFatal("sync_cloudbrain_status", &BaseConfig{
+		Enabled:    true,
+		RunAtStart: false,
+		Schedule:   "@every 10m",
+	}, func(ctx context.Context, _ *models.User, _ Config) error {
+		repo.SyncCloudbrainStatus()
+		return nil
+	})
+}
+
 func initBasicTasks() {
 	registerUpdateMirrorTask()
 	registerRepoHealthCheck()
@@ -202,4 +213,6 @@ func initBasicTasks() {
 
 	registerHandleRepoAndUserStatistic()
 	registerHandleSummaryStatistic()
+
+	registerSyncCloudbrainStatus()
 }

modules/storage/storage.go

@@ -51,6 +51,7 @@ func Init() error {
 	switch setting.Attachment.StoreType {
 	case LocalStorageType:
 		Attachments, err = NewLocalStorage(setting.Attachment.Path)
+		log.Info("local storage inited.")
 	case MinioStorageType:
 		minio := setting.Attachment.Minio
 		Attachments, err = NewMinioStorage(
@@ -62,6 +63,7 @@ func Init() error {
 			minio.BasePath,
 			minio.UseSSL,
 		)
+		log.Info("minio storage inited.")
 	default:
 		return fmt.Errorf("Unsupported attachment store type: %s", setting.Attachment.StoreType)
 	}
@@ -71,6 +73,7 @@ func Init() error {
 		log.Error("obs.New failed:", err)
 		return err
 	}
+	log.Info("obs cli inited.")
 
 	if err != nil {
 		return err

routers/repo/ai_model_manage.go

@@ -25,7 +25,7 @@ const (
 	MODEL_NOT_LATEST       = 0
 )
 
-func saveModelByParameters(jobId string, versionName string, name string, version string, label string, description string, userId int64) error {
+func saveModelByParameters(jobId string, versionName string, name string, version string, label string, description string, userId int64, userName string) error {
 	aiTask, err := models.GetCloudbrainByJobIDAndVersionName(jobId, versionName)
 	//aiTask, err := models.GetCloudbrainByJobID(jobId)
 	if err != nil {
@@ -82,6 +82,7 @@ func saveModelByParameters(jobId string, versionName string, name string, versio
 		AttachmentId:  aiTask.Uuid,
 		RepoId:        aiTask.RepoID,
 		UserId:        userId,
+		UserName:      userName,
 		CodeBranch:    aiTask.BranchName,
 		CodeCommitID:  aiTask.CommitID,
 		Engine:        aiTask.EngineID,
@@ -122,7 +123,7 @@ func SaveModel(ctx *context.Context) {
 		return
 	}
 
-	err := saveModelByParameters(JobId, VersionName, name, version, label, description, ctx.User.ID)
+	err := saveModelByParameters(JobId, VersionName, name, version, label, description, ctx.User.ID, ctx.User.Name)
 
 	if err != nil {
 		log.Info("save model error." + err.Error())

routers/repo/cloudbrain.go

@@ -715,3 +715,78 @@ func downloadRateCode(repo *models.Repository, taskName, gitPath, codePath, benc
 
 	return nil
 }
+
+func SyncCloudbrainStatus() {
+	cloudBrains, err := models.GetCloudBrainUnStoppedJob()
+	if err != nil {
+		log.Error("GetCloudBrainUnStoppedJob failed:", err.Error())
+		return
+	}
+
+	for _, task := range cloudBrains {
+		if task.Type == models.TypeCloudBrainOne {
+			result, err := cloudbrain.GetJob(task.JobID)
+			if err != nil {
+				log.Error("GetJob(%s) failed:%v", task.JobName, err)
+				continue
+			}
+
+			if result != nil {
+				jobRes, _ := models.ConvertToJobResultPayload(result.Payload)
+				taskRoles := jobRes.TaskRoles
+				taskRes, _ := models.ConvertToTaskPod(taskRoles[cloudbrain.SubTaskName].(map[string]interface{}))
+				task.Status = taskRes.TaskStatuses[0].State
+				if task.Status != string(models.JobWaiting) {
+					err = models.UpdateJob(task)
+					if err != nil {
+						log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
+						continue
+					}
+				}
+			}
+		} else if task.Type == models.TypeCloudBrainTwo {
+			if task.JobType == string(models.JobTypeDebug) {
+				result, err := modelarts.GetJob(task.JobID)
+				if err != nil {
+					log.Error("GetJob(%s) failed:%v", task.JobName, err)
+					continue
+				}
+
+				if result != nil {
+					task.Status = result.Status
+
+					err = models.UpdateJob(task)
+					if err != nil {
+						log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
+						continue
+					}
+				}
+			} else if task.JobType == string(models.JobTypeTrain) {
+				result, err := modelarts.GetTrainJob(task.JobID, strconv.FormatInt(task.VersionID, 10))
+				if err != nil {
+					log.Error("GetTrainJob(%s) failed:%v", task.JobName, err)
+					continue
+				}
+
+				if result != nil {
+					task.Status = modelarts.TransTrainJobStatus(result.IntStatus)
+					task.Duration = result.Duration
+					task.TrainJobDuration = result.TrainJobDuration
+
+					err = models.UpdateJob(task)
+					if err != nil {
+						log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
+						continue
+					}
+				}
+			} else {
+				log.Error("task.JobType(%s) is error:%s", task.JobName, task.JobType)
+			}
+
+		} else {
+			log.Error("task.Type(%s) is error:%d", task.JobName, task.Type)
+		}
+	}
+
+	return
+}

routers/repo/modelarts.go

@@ -364,13 +364,17 @@ func trainJobNewDataPrepare(ctx *context.Context) error {
 
 	Branches, err := ctx.Repo.GitRepo.GetBranches()
 	if err != nil {
+		log.Error("GetBranches failed:%v", err)
 		ctx.ServerError("GetBranches error:", err)
 		return err
 	}
-	ctx.Data["Branches"] = Branches
+	if Branches != nil {
+		ctx.Data["Branches"] = Branches
+	}
+
 	ctx.Data["BranchesCount"] = len(Branches)
 	ctx.Data["params"] = ""
-	ctx.Data["BranchName"] = ctx.Repo.BranchName
+	ctx.Data["branchName"] = ctx.Repo.BranchName
 
 	configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
 	if err != nil {

templates/repo/modelarts/trainjob/new.tmpl

@@ -103,9 +103,9 @@
                                     {{end}}
                                 {{end}}
                             {{else}}
-                                <option name="branch_name" value="{{.BranchName}}">{{.BranchName}}</option>
+                                <option name="branch_name" value="{{.branchName}}">{{.branchName}}</option>
                                 {{range $k, $v :=.Branches}}
-                                    {{ if ne $v $.BranchName }}
+                                    {{ if ne $v $.branchName }}
                                     <option name="branch_name" value="{{$v}}">{{$v}}</option>
                                     {{end}}
                                 {{end}}