From 79cc83e9c927c721b7e72d23fa07dbf108f4edb2 Mon Sep 17 00:00:00 2001 From: lewis <747342561@qq.com> Date: Fri, 22 Jul 2022 17:08:51 +0800 Subject: [PATCH] sync modelarts --- models/cloudbrain.go | 72 +++- models/cloudbrain_temp.go | 57 +++ modules/cloudbrain/cloudbrain.go | 8 +- modules/modelarts/modelarts.go | 716 ++++++++++++++++++++----------- modules/modelarts/resty.go | 192 ++++++++- routers/api/v1/repo/modelarts.go | 127 +----- routers/repo/cloudbrain.go | 58 +-- routers/repo/grampus.go | 4 +- routers/repo/modelarts.go | 317 ++++++-------- routers/routes/routes.go | 3 +- 10 files changed, 921 insertions(+), 633 deletions(-) create mode 100755 models/cloudbrain_temp.go diff --git a/models/cloudbrain.go b/models/cloudbrain.go index f32ad5870..7ecf2670b 100755 --- a/models/cloudbrain.go +++ b/models/cloudbrain.go @@ -31,9 +31,11 @@ const ( ) const ( - NPUResource = "NPU" - GPUResource = "CPU/GPU" - AllResource = "all" + TempJobIdPrefix = "TEMP" + JobStatusTemp = "TEMP" + NPUResource = "NPU" + GPUResource = "CPU/GPU" + AllResource = "all" //notebook storage category EVSCategory = "EVS" @@ -353,6 +355,7 @@ type CloudbrainsOptions struct { RepoID int64 // include all repos if empty UserID int64 JobID string + JobName string SortType string CloudbrainIDs []int64 JobStatus []string @@ -1256,6 +1259,52 @@ type LogFile struct { Name string } +type JobList struct { + JobName string `json:"job_name"` + JobID int64 `json:"job_id"` + VersionID int64 `json:"version_id"` + VersionCount int64 `json:"version_count"` + Description string `json:"job_desc"` + IntStatus int `json:"status"` +} + +type GetTrainJobListResult struct { + ErrorResult + JobTotalCount int `json:"job_total_count"` //查询到的用户创建作业总数 + JobCountLimit int `json:"job_count_limit"` //用户还可以创建训练作业的数量 + Quotas int `json:"quotas"` //训练作业的运行数量上限 + JobList []JobList `json:"jobs"` +} + +type JobVersionList struct { + VersionName string `json:"version_name"` + VersionID int64 `json:"version_id"` + IntStatus int `json:"status"` +} + +type GetTrainJobVersionListResult struct { + ErrorResult + JobID int64 `json:"job_id"` + JobName string `json:"job_name"` + JobDesc string `json:"job_desc"` + VersionCount int64 `json:"version_count"` + JobVersionList []JobVersionList `json:"versions"` +} + +type NotebookList struct { + JobName string `json:"name"` + JobID string `json:"id"` + Status string `json:"status"` +} + +type GetNotebookListResult struct { + TotalCount int64 `json:"total"` //总的记录数量 + CurrentPage int `json:"current"` //当前页数 + TotalPages int `json:"pages"` //总的页数 + Size int `json:"size"` //每一页的数量 + NotebookList []NotebookList `json:"data"` +} + //Grampus type GrampusResult struct { ErrorCode int `json:"errorCode"` @@ -1568,6 +1617,12 @@ func CloudbrainsVersionList(opts *CloudbrainsOptions) ([]*CloudbrainInfo, int, e ) } + if (opts.JobName) != "" { + cond = cond.And( + builder.Eq{"cloudbrain.job_name": opts.JobName}, + ) + } + if len(opts.JobTypes) > 0 { cond = cond.And( builder.In("cloudbrain.job_type", opts.JobTypes), @@ -1701,9 +1756,9 @@ func SetTrainJobStatusByJobID(jobID string, status string, duration int64, train return } -func SetVersionCountAndLatestVersion(jobID string, versionName string, versionCount int, isLatestVersion string, totalVersionCount int) (err error) { - cb := &Cloudbrain{JobID: jobID, VersionName: versionName, VersionCount: versionCount, IsLatestVersion: isLatestVersion, TotalVersionCount: totalVersionCount} - _, err = x.Cols("version_Count", "is_latest_version", "total_version_count").Where("cloudbrain.job_id=? AND cloudbrain.version_name=?", jobID, versionName).Update(cb) +func SetVersionCountAndLatestVersion(jobName string, versionName string, versionCount int, isLatestVersion string, totalVersionCount int) (err error) { + cb := &Cloudbrain{JobName: jobName, VersionName: versionName, VersionCount: versionCount, IsLatestVersion: isLatestVersion, TotalVersionCount: totalVersionCount} + _, err = x.Cols("version_Count", "is_latest_version", "total_version_count").Where("cloudbrain.job_name=? AND cloudbrain.version_name=?", jobName, versionName).Update(cb) return } @@ -2123,3 +2178,8 @@ func GetCloudbrainByIDs(ids []int64) ([]*Cloudbrain, error) { In("id", ids). Find(&cloudbrains) } + +func GetCloudbrainCountByJobName(jobName, jobType string) (int, error) { + count, err := x.Where("job_name = ? and job_type= ?", jobName, jobType).Count(new(Cloudbrain)) + return int(count), err +} diff --git a/models/cloudbrain_temp.go b/models/cloudbrain_temp.go new file mode 100755 index 000000000..e7d424006 --- /dev/null +++ b/models/cloudbrain_temp.go @@ -0,0 +1,57 @@ +package models + +import ( + "time" + + "code.gitea.io/gitea/modules/timeutil" +) + +const ( +//TempJobIdPrefix = "TEMP" + +) + +type CloudbrainTemp struct { + CloudbrainID int64 `xorm:"pk"` + JobName string + Type int + JobType string `xorm:"INDEX NOT NULL DEFAULT 'DEBUG'"` + Status string `xorm:"INDEX NOT NULL DEFAULT 'TEMP'"` + VersionCount int `xorm:"NOT NULL DEFAULT 0"` + QueryTimes int `xorm:"INDEX NOT NULL DEFAULT 0"` + CreatedUnix timeutil.TimeStamp `xorm:"INDEX"` + UpdatedUnix timeutil.TimeStamp `xorm:"INDEX updated"` + DeletedAt time.Time `xorm:"deleted"` +} + +func InsertCloudbrainTemp(temp *CloudbrainTemp) (err error) { + if _, err = x.Insert(temp); err != nil { + return err + } + + return nil +} + +func getCloudBrainTemp(temp *CloudbrainTemp) (*CloudbrainTemp, error) { + has, err := x.Get(temp) + if err != nil { + return nil, err + } else if !has { + return nil, ErrJobNotExist{} + } + return temp, nil +} + +func GetCloudbrainTempByCloudbrainID(id int64) (*CloudbrainTemp, error) { + temp := &CloudbrainTemp{CloudbrainID: id} + return getCloudBrainTemp(temp) +} + +func DeleteCloudbrainTemp(temp *CloudbrainTemp) error { + return deleteCloudbrainTemp(x, temp) +} + +func deleteCloudbrainTemp(e Engine, temp *CloudbrainTemp) error { + _, err := e.Where("cloudbrain_id = ?", temp.CloudbrainID).Delete(temp) + return err +} diff --git a/modules/cloudbrain/cloudbrain.go b/modules/cloudbrain/cloudbrain.go index 6cbb97999..5b592a11b 100755 --- a/modules/cloudbrain/cloudbrain.go +++ b/modules/cloudbrain/cloudbrain.go @@ -142,8 +142,8 @@ func isAdminOrImageCreater(ctx *context.Context, image *models.Image, err error) func AdminOrOwnerOrJobCreaterRight(ctx *context.Context) { - var ID = ctx.Params(":id") - job, err := models.GetCloudbrainByID(ID) + var id = ctx.Params(":id") + job, err := models.GetCloudbrainByID(id) if err != nil { log.Error("GetCloudbrainByID failed:%v", err.Error()) ctx.NotFound(ctx.Req.URL.RequestURI(), nil) @@ -158,8 +158,8 @@ func AdminOrOwnerOrJobCreaterRight(ctx *context.Context) { func AdminOrJobCreaterRight(ctx *context.Context) { - var ID = ctx.Params(":id") - job, err := models.GetCloudbrainByID(ID) + var id = ctx.Params(":id") + job, err := models.GetCloudbrainByID(id) if err != nil { log.Error("GetCloudbrainByID failed:%v", err.Error()) ctx.NotFound(ctx.Req.URL.RequestURI(), nil) diff --git a/modules/modelarts/modelarts.go b/modules/modelarts/modelarts.go index 1f39d0fac..ab5070262 100755 --- a/modules/modelarts/modelarts.go +++ b/modules/modelarts/modelarts.go @@ -4,8 +4,11 @@ import ( "encoding/json" "errors" "fmt" + "math/rand" "path" "strconv" + "strings" + "time" "code.gitea.io/gitea/modules/timeutil" @@ -59,7 +62,7 @@ const ( PerPage = 10 IsLatestVersion = "1" NotLatestVersion = "0" - VersionCount = 1 + VersionCountOne = 1 SortByCreateTime = "create_time" ConfigTypeCustom = "custom" @@ -264,31 +267,13 @@ func GenerateNotebook2(ctx *context.Context, displayJobName, jobName, uuid, desc log.Error("GetNotebookImageName failed: %v", err.Error()) return err } + createTime := timeutil.TimeStampNow() - jobResult, err := createNotebook2(models.CreateNotebook2Params{ - JobName: jobName, - Description: description, - Flavor: flavor, - Duration: autoStopDurationMs, - ImageID: imageId, - PoolID: poolInfos.PoolInfo[0].PoolId, - Feature: models.NotebookFeature, - Volume: models.VolumeReq{ - Capacity: setting.Capacity, - Category: models.EVSCategory, - Ownership: models.ManagedOwnership, - }, - WorkspaceID: "0", - }) - if err != nil { - log.Error("createNotebook2 failed: %v", err.Error()) - return err - } - err = models.CreateCloudbrain(&models.Cloudbrain{ - Status: jobResult.Status, + task := &models.Cloudbrain{ + Status: string(models.ModelArtsTrainJobWaiting), UserID: ctx.User.ID, RepoID: ctx.Repo.Repository.ID, - JobID: jobResult.ID, + JobID: models.TempJobIdPrefix + jobName + strconv.Itoa(int(rand.New(rand.NewSource(time.Now().UnixNano())).Int31n(100000))), JobName: jobName, FlavorCode: flavor, DisplayJobName: displayJobName, @@ -300,16 +285,66 @@ func GenerateNotebook2(ctx *context.Context, displayJobName, jobName, uuid, desc Description: description, CreatedUnix: createTime, UpdatedUnix: createTime, - }) + } + err = models.CreateCloudbrain(task) if err != nil { + log.Error("CreateCloudbrain(%s) failed:%v", displayJobName, err.Error()) return err } - task, err := models.GetCloudbrainByName(jobName) + + jobResult, err := createNotebook2(models.CreateNotebook2Params{ + JobName: jobName, + Description: description, + Flavor: flavor, + Duration: autoStopDurationMs, + ImageID: imageId, + PoolID: poolInfos.PoolInfo[0].PoolId, + Feature: models.NotebookFeature, + Volume: models.VolumeReq{ + Capacity: setting.Capacity, + Category: models.EVSCategory, + Ownership: models.ManagedOwnership, + }, + WorkspaceID: "0", + }) if err != nil { - log.Error("GetCloudbrainByName failed: %v", err.Error()) - return err + log.Error("createNotebook2 failed: %v", err.Error()) + if strings.HasPrefix(err.Error(), UnknownErrorPrefix) { + log.Info("(%s)unknown error, set temp status", displayJobName) + errTemp := models.InsertCloudbrainTemp(&models.CloudbrainTemp{ + CloudbrainID: task.ID, + Status: models.JobStatusTemp, + Type: task.Type, + JobName: task.JobName, + JobType: task.JobType, + }) + if errTemp != nil { + log.Error("InsertCloudbrainTemp failed: %v", errTemp.Error()) + return errTemp + } + } else { + task.Status = string(models.ModelArtsCreateFailed) + errTemp := models.UpdateJob(task) + if errTemp != nil { + log.Error("UpdateJob failed: %v", errTemp.Error()) + } + errTemp = models.DeleteJob(task) + if errTemp != nil { + log.Error("DeleteJob failed: %v", errTemp.Error()) + } + return err + } + } else { + task.Status = jobResult.Status + task.JobID = jobResult.ID + err = models.UpdateJob(task) + if err != nil { + log.Error("UpdateJob failed: %v", err.Error()) + return err + } } + stringId := strconv.FormatInt(task.ID, 10) notification.NotifyOtherTask(ctx.User, ctx.Repo.Repository, stringId, displayJobName, models.ActionCreateDebugNPUTask) return nil @@ -317,66 +352,15 @@ func GenerateNotebook2(ctx *context.Context, displayJobName, jobName, uuid, desc func GenerateTrainJob(ctx *context.Context, req *GenerateTrainJobReq) (err error) { createTime := timeutil.TimeStampNow() - var jobResult *models.CreateTrainJobResult - var createErr error - if req.EngineID < 0 { - jobResult, createErr = createTrainJobUserImage(models.CreateUserImageTrainJobParams{ - JobName: req.JobName, - Description: req.Description, - Config: models.UserImageConfig{ - WorkServerNum: req.WorkServerNumber, - AppUrl: req.CodeObsPath, - BootFileUrl: req.BootFileUrl, - DataUrl: req.DataUrl, - TrainUrl: req.TrainUrl, - LogUrl: req.LogUrl, - PoolID: req.PoolID, - CreateVersion: true, - Flavor: models.Flavor{ - Code: req.FlavorCode, - }, - Parameter: req.Parameters, - UserImageUrl: req.UserImageUrl, - UserCommand: req.UserCommand, - }, - }) - } else { - jobResult, createErr = createTrainJob(models.CreateTrainJobParams{ - JobName: req.JobName, - Description: req.Description, - Config: models.Config{ - WorkServerNum: req.WorkServerNumber, - AppUrl: req.CodeObsPath, - BootFileUrl: req.BootFileUrl, - 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 createErr != nil { - log.Error("CreateJob failed: %v", createErr.Error()) - return createErr - } - jobId := strconv.FormatInt(jobResult.JobID, 10) - createErr = models.CreateCloudbrain(&models.Cloudbrain{ - Status: TransTrainJobStatus(jobResult.Status), + task := &models.Cloudbrain{ + Status: string(models.ModelArtsTrainJobWaiting), UserID: ctx.User.ID, RepoID: ctx.Repo.Repository.ID, - JobID: jobId, + JobID: models.TempJobIdPrefix + req.JobName + strconv.Itoa(int(rand.New(rand.NewSource(time.Now().UnixNano())).Int31n(100000))), JobName: req.JobName, DisplayJobName: req.DisplayJobName, JobType: string(models.JobTypeTrain), Type: models.TypeCloudBrainTwo, - VersionID: jobResult.VersionID, - VersionName: jobResult.VersionName, Uuid: req.Uuid, DatasetName: req.DatasetName, CommitID: req.CommitID, @@ -398,49 +382,21 @@ func GenerateTrainJob(ctx *context.Context, req *GenerateTrainJobReq) (err error TotalVersionCount: req.TotalVersionCount, CreatedUnix: createTime, UpdatedUnix: createTime, - }) - - if createErr != nil { - log.Error("CreateCloudbrain(%s) failed:%v", req.DisplayJobName, createErr.Error()) - return createErr } - notification.NotifyOtherTask(ctx.User, ctx.Repo.Repository, jobId, req.DisplayJobName, models.ActionCreateTrainTask) - return nil -} - -func GenerateModelConvertTrainJob(req *GenerateTrainJobReq) (*models.CreateTrainJobResult, error) { - - return createTrainJobUserImage(models.CreateUserImageTrainJobParams{ - JobName: req.JobName, - Description: req.Description, - Config: models.UserImageConfig{ - WorkServerNum: req.WorkServerNumber, - AppUrl: req.CodeObsPath, - BootFileUrl: req.BootFileUrl, - DataUrl: req.DataUrl, - TrainUrl: req.TrainUrl, - LogUrl: req.LogUrl, - PoolID: req.PoolID, - CreateVersion: true, - Flavor: models.Flavor{ - Code: req.FlavorCode, - }, - Parameter: req.Parameters, - UserImageUrl: req.UserImageUrl, - UserCommand: req.UserCommand, - }, - }) -} + err = models.CreateCloudbrain(task) + if err != nil { + log.Error("CreateCloudbrain(%s) failed:%v", req.DisplayJobName, err.Error()) + return err + } -func GenerateTrainJobVersion(ctx *context.Context, req *GenerateTrainJobReq, jobId string) (err error) { - createTime := timeutil.TimeStampNow() var jobResult *models.CreateTrainJobResult var createErr error - log.Info(" req.EngineID =" + fmt.Sprint(req.EngineID)) + if req.EngineID < 0 { - jobResult, createErr = createTrainJobVersionUserImage(models.CreateTrainJobVersionUserImageParams{ + jobResult, createErr = createTrainJobUserImage(models.CreateUserImageTrainJobParams{ + JobName: req.JobName, Description: req.Description, - Config: models.TrainJobVersionUserImageConfig{ + Config: models.UserImageConfig{ WorkServerNum: req.WorkServerNumber, AppUrl: req.CodeObsPath, BootFileUrl: req.BootFileUrl, @@ -448,19 +404,20 @@ func GenerateTrainJobVersion(ctx *context.Context, req *GenerateTrainJobReq, job TrainUrl: req.TrainUrl, LogUrl: req.LogUrl, PoolID: req.PoolID, + CreateVersion: true, Flavor: models.Flavor{ Code: req.FlavorCode, }, Parameter: req.Parameters, - PreVersionId: req.PreVersionId, UserImageUrl: req.UserImageUrl, UserCommand: req.UserCommand, }, - }, jobId) + }) } else { - jobResult, createErr = createTrainJobVersion(models.CreateTrainJobVersionParams{ + jobResult, createErr = createTrainJob(models.CreateTrainJobParams{ + JobName: req.JobName, Description: req.Description, - Config: models.TrainJobVersionConfig{ + Config: models.Config{ WorkServerNum: req.WorkServerNumber, AppUrl: req.CodeObsPath, BootFileUrl: req.BootFileUrl, @@ -469,87 +426,60 @@ func GenerateTrainJobVersion(ctx *context.Context, req *GenerateTrainJobReq, job TrainUrl: req.TrainUrl, LogUrl: req.LogUrl, PoolID: req.PoolID, + CreateVersion: true, Flavor: models.Flavor{ Code: req.FlavorCode, }, - Parameter: req.Parameters, - PreVersionId: req.PreVersionId, + Parameter: req.Parameters, }, - }, jobId) - } - if createErr != nil { - log.Error("CreateJob failed: %v", createErr.Error()) - return createErr - } - - var jobTypes []string - jobTypes = append(jobTypes, string(models.JobTypeTrain)) - repo := ctx.Repo.Repository - VersionTaskList, VersionListCount, createErr := models.CloudbrainsVersionList(&models.CloudbrainsOptions{ - RepoID: repo.ID, - Type: models.TypeCloudBrainTwo, - JobTypes: jobTypes, - JobID: strconv.FormatInt(jobResult.JobID, 10), - }) - if createErr != nil { - ctx.ServerError("Cloudbrain", createErr) - return createErr - } - //将当前版本的isLatestVersion设置为"1"和任务数量更新,任务数量包括当前版本数VersionCount和历史创建的总版本数TotalVersionCount - - createErr = 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, - DisplayJobName: req.DisplayJobName, - JobType: string(models.JobTypeTrain), - Type: models.TypeCloudBrainTwo, - VersionID: jobResult.VersionID, - VersionName: jobResult.VersionName, - Uuid: req.Uuid, - DatasetName: req.DatasetName, - CommitID: req.CommitID, - IsLatestVersion: req.IsLatestVersion, - PreVersionName: req.PreVersionName, - ComputeResource: models.NPUResource, - EngineID: req.EngineID, - TrainUrl: req.TrainUrl, - BranchName: req.BranchName, - Parameters: req.Params, - BootFile: req.BootFile, - DataUrl: req.DataUrl, - LogUrl: req.LogUrl, - PreVersionId: req.PreVersionId, - FlavorCode: req.FlavorCode, - Description: req.Description, - WorkServerNumber: req.WorkServerNumber, - FlavorName: req.FlavorName, - EngineName: req.EngineName, - TotalVersionCount: VersionTaskList[0].TotalVersionCount + 1, - VersionCount: VersionListCount + 1, - CreatedUnix: createTime, - UpdatedUnix: createTime, - }) - if createErr != nil { - log.Error("CreateCloudbrain(%s) failed:%v", req.JobName, createErr.Error()) - return createErr + }) } - - //将训练任务的上一版本的isLatestVersion设置为"0" - createErr = models.SetVersionCountAndLatestVersion(strconv.FormatInt(jobResult.JobID, 10), VersionTaskList[0].VersionName, VersionCount, NotLatestVersion, TotalVersionCount) if createErr != nil { - ctx.ServerError("Update IsLatestVersion failed", createErr) - return createErr + log.Error("createTrainJob failed: %v", createErr.Error()) + if strings.HasPrefix(createErr.Error(), UnknownErrorPrefix) { + log.Info("(%s)unknown error, set temp status", req.DisplayJobName) + errTemp := models.InsertCloudbrainTemp(&models.CloudbrainTemp{ + CloudbrainID: task.ID, + Status: models.JobStatusTemp, + Type: task.Type, + JobName: task.JobName, + JobType: task.JobType, + }) + if errTemp != nil { + log.Error("InsertCloudbrainTemp failed: %v", errTemp.Error()) + return errTemp + } + } else { + task.Status = string(models.ModelArtsTrainJobFailed) + errTemp := models.UpdateJob(task) + if errTemp != nil { + log.Error("UpdateJob failed: %v", errTemp.Error()) + } + errTemp = models.DeleteJob(task) + if errTemp != nil { + log.Error("DeleteJob failed: %v", errTemp.Error()) + } + return createErr + } + } else { + task.Status = TransTrainJobStatus(jobResult.Status) + task.JobID = strconv.FormatInt(jobResult.JobID, 10) + task.VersionID = jobResult.VersionID + task.VersionName = jobResult.VersionName + err = models.UpdateJob(task) + if err != nil { + log.Error("UpdateJob failed: %v", err.Error()) + return err + } } - return createErr + notification.NotifyOtherTask(ctx.User, ctx.Repo.Repository, task.JobID, req.DisplayJobName, models.ActionCreateTrainTask) + return nil } -func GenerateTrainJobVersionByUserImage(ctx *context.Context, req *GenerateTrainJobReq, jobId string) (err error) { - createTime := timeutil.TimeStampNow() - jobResult, err := createTrainJobUserImage(models.CreateUserImageTrainJobParams{ +func GenerateModelConvertTrainJob(req *GenerateTrainJobReq) (*models.CreateTrainJobResult, error) { + + return createTrainJobUserImage(models.CreateUserImageTrainJobParams{ JobName: req.JobName, Description: req.Description, Config: models.UserImageConfig{ @@ -569,11 +499,9 @@ func GenerateTrainJobVersionByUserImage(ctx *context.Context, req *GenerateTrain UserCommand: req.UserCommand, }, }) - if err != nil { - log.Error("CreateJob failed: %v", err.Error()) - return err - } +} +func GenerateTrainJobVersion(ctx *context.Context, req *GenerateTrainJobReq, jobId string) (err error) { var jobTypes []string jobTypes = append(jobTypes, string(models.JobTypeTrain)) repo := ctx.Repo.Repository @@ -581,7 +509,7 @@ func GenerateTrainJobVersionByUserImage(ctx *context.Context, req *GenerateTrain RepoID: repo.ID, Type: models.TypeCloudBrainTwo, JobTypes: jobTypes, - JobID: strconv.FormatInt(jobResult.JobID, 10), + JobID: jobId, }) if err != nil { ctx.ServerError("Cloudbrain", err) @@ -589,25 +517,23 @@ func GenerateTrainJobVersionByUserImage(ctx *context.Context, req *GenerateTrain } //将当前版本的isLatestVersion设置为"1"和任务数量更新,任务数量包括当前版本数VersionCount和历史创建的总版本数TotalVersionCount - err = models.CreateCloudbrain(&models.Cloudbrain{ - Status: TransTrainJobStatus(jobResult.Status), + createTime := timeutil.TimeStampNow() + task := &models.Cloudbrain{ + Status: models.JobStatusTemp, UserID: ctx.User.ID, RepoID: ctx.Repo.Repository.ID, - JobID: strconv.FormatInt(jobResult.JobID, 10), + JobID: jobId, JobName: req.JobName, DisplayJobName: req.DisplayJobName, JobType: string(models.JobTypeTrain), Type: models.TypeCloudBrainTwo, - VersionID: jobResult.VersionID, - VersionName: jobResult.VersionName, Uuid: req.Uuid, DatasetName: req.DatasetName, CommitID: req.CommitID, IsLatestVersion: req.IsLatestVersion, PreVersionName: req.PreVersionName, ComputeResource: models.NPUResource, - EngineID: MORDELART_USER_IMAGE_ENGINE_ID, - Image: req.UserImageUrl, + EngineID: req.EngineID, TrainUrl: req.TrainUrl, BranchName: req.BranchName, Parameters: req.Params, @@ -624,20 +550,103 @@ func GenerateTrainJobVersionByUserImage(ctx *context.Context, req *GenerateTrain VersionCount: VersionListCount + 1, CreatedUnix: createTime, UpdatedUnix: createTime, - }) + } + err = models.CreateCloudbrain(task) if err != nil { log.Error("CreateCloudbrain(%s) failed:%v", req.JobName, err.Error()) return err } //将训练任务的上一版本的isLatestVersion设置为"0" - err = models.SetVersionCountAndLatestVersion(strconv.FormatInt(jobResult.JobID, 10), VersionTaskList[0].VersionName, VersionCount, NotLatestVersion, TotalVersionCount) + err = models.SetVersionCountAndLatestVersion(req.JobName, VersionTaskList[0].VersionName, VersionListCount, NotLatestVersion, VersionTaskList[0].TotalVersionCount) if err != nil { ctx.ServerError("Update IsLatestVersion failed", err) return err } - return err + var jobResult *models.CreateTrainJobResult + var createErr error + + if req.EngineID < 0 { + jobResult, createErr = createTrainJobVersionUserImage(models.CreateTrainJobVersionUserImageParams{ + Description: req.Description, + Config: models.TrainJobVersionUserImageConfig{ + WorkServerNum: req.WorkServerNumber, + AppUrl: req.CodeObsPath, + BootFileUrl: req.BootFileUrl, + DataUrl: req.DataUrl, + TrainUrl: req.TrainUrl, + LogUrl: req.LogUrl, + PoolID: req.PoolID, + Flavor: models.Flavor{ + Code: req.FlavorCode, + }, + Parameter: req.Parameters, + PreVersionId: req.PreVersionId, + UserImageUrl: req.UserImageUrl, + UserCommand: req.UserCommand, + }, + }, jobId) + } else { + jobResult, createErr = createTrainJobVersion(models.CreateTrainJobVersionParams{ + Description: req.Description, + Config: models.TrainJobVersionConfig{ + WorkServerNum: req.WorkServerNumber, + AppUrl: req.CodeObsPath, + BootFileUrl: req.BootFileUrl, + DataUrl: req.DataUrl, + EngineID: req.EngineID, + TrainUrl: req.TrainUrl, + LogUrl: req.LogUrl, + PoolID: req.PoolID, + Flavor: models.Flavor{ + Code: req.FlavorCode, + }, + Parameter: req.Parameters, + PreVersionId: req.PreVersionId, + }, + }, jobId) + } + if createErr != nil { + log.Error("createTrainJobVersion failed: %v", err.Error()) + if strings.HasPrefix(err.Error(), UnknownErrorPrefix) { + log.Info("(%s)unknown error, set temp status", req.DisplayJobName) + errTemp := models.InsertCloudbrainTemp(&models.CloudbrainTemp{ + CloudbrainID: task.ID, + Status: models.JobStatusTemp, + Type: task.Type, + JobName: task.JobName, + JobType: task.JobType, + }) + if errTemp != nil { + log.Error("InsertCloudbrainTemp failed: %v", errTemp.Error()) + return errTemp + } + } else { + task.Status = string(models.ModelArtsTrainJobFailed) + errTemp := models.UpdateJob(task) + if errTemp != nil { + log.Error("UpdateJob failed: %v", errTemp.Error()) + } + errTemp = models.DeleteJob(task) + if errTemp != nil { + log.Error("DeleteJob failed: %v", errTemp.Error()) + } + return createErr + } + } else { + task.Status = TransTrainJobStatus(jobResult.Status) + task.JobID = strconv.FormatInt(jobResult.JobID, 10) + task.VersionID = jobResult.VersionID + task.VersionName = jobResult.VersionName + err = models.UpdateJob(task) + if err != nil { + log.Error("UpdateJob failed: %v", err.Error()) + return err + } + } + + return nil } func TransTrainJobStatus(status int) string { @@ -700,47 +709,22 @@ func GetOutputPathByCount(TotalVersionCount int) (VersionOutputPath string) { func GenerateInferenceJob(ctx *context.Context, req *GenerateInferenceJobReq) (err error) { createTime := timeutil.TimeStampNow() - jobResult, err := createInferenceJob(models.CreateInferenceJobParams{ - JobName: req.JobName, - Description: req.Description, - InfConfig: models.InfConfig{ - WorkServerNum: req.WorkServerNumber, - AppUrl: req.CodeObsPath, - BootFileUrl: req.BootFileUrl, - 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()) + log.Error("GetAttachmentByUUID(%s) failed:%v", req.DisplayJobName, err.Error()) return err } - jobID := strconv.FormatInt(jobResult.JobID, 10) - err = models.CreateCloudbrain(&models.Cloudbrain{ - Status: TransTrainJobStatus(jobResult.Status), + + task := &models.Cloudbrain{ + Status: string(models.ModelArtsTrainJobWaiting), UserID: ctx.User.ID, RepoID: ctx.Repo.Repository.ID, - JobID: jobID, + JobID: models.TempJobIdPrefix + req.JobName + strconv.Itoa(int(rand.New(rand.NewSource(time.Now().UnixNano())).Int31n(100000))), JobName: req.JobName, DisplayJobName: req.DisplayJobName, JobType: string(models.JobTypeInference), Type: models.TypeCloudBrainTwo, - VersionID: jobResult.VersionID, - VersionName: jobResult.VersionName, Uuid: req.Uuid, DatasetName: attach.Name, CommitID: req.CommitID, @@ -767,13 +751,74 @@ func GenerateInferenceJob(ctx *context.Context, req *GenerateInferenceJobReq) (e ResultUrl: req.ResultUrl, CreatedUnix: createTime, UpdatedUnix: createTime, - }) + } + err = models.CreateCloudbrain(task) if err != nil { log.Error("CreateCloudbrain(%s) failed:%v", req.JobName, err.Error()) return err } - notification.NotifyOtherTask(ctx.User, ctx.Repo.Repository, jobID, req.DisplayJobName, models.ActionCreateInferenceTask) + + jobResult, err := createInferenceJob(models.CreateInferenceJobParams{ + JobName: req.JobName, + Description: req.Description, + InfConfig: models.InfConfig{ + WorkServerNum: req.WorkServerNumber, + AppUrl: req.CodeObsPath, + BootFileUrl: req.BootFileUrl, + 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("createTrainJob failed: %v", err.Error()) + if strings.HasPrefix(err.Error(), UnknownErrorPrefix) { + log.Info("(%s)unknown error, set temp status", req.DisplayJobName) + err = models.InsertCloudbrainTemp(&models.CloudbrainTemp{ + CloudbrainID: task.ID, + Status: models.JobStatusTemp, + Type: task.Type, + JobName: task.JobName, + JobType: task.JobType, + }) + if err != nil { + log.Error("InsertCloudbrainTemp failed: %v", err.Error()) + return err + } + } else { + task.Status = string(models.ModelArtsTrainJobFailed) + errTemp := models.UpdateJob(task) + if errTemp != nil { + log.Error("UpdateJob failed: %v", errTemp.Error()) + } + errTemp = models.DeleteJob(task) + if errTemp != nil { + log.Error("DeleteJob failed: %v", errTemp.Error()) + } + return err + } + } else { + task.Status = TransTrainJobStatus(jobResult.Status) + task.JobID = strconv.FormatInt(jobResult.JobID, 10) + task.VersionID = jobResult.VersionID + task.VersionName = jobResult.VersionName + err = models.UpdateJob(task) + if err != nil { + log.Error("UpdateJob failed: %v", err.Error()) + return err + } + } + + notification.NotifyOtherTask(ctx.User, ctx.Repo.Repository, task.JobID, req.DisplayJobName, models.ActionCreateInferenceTask) + return nil } @@ -799,3 +844,194 @@ func GetNotebookImageName(imageId string) (string, error) { return imageName, nil } + +func HandleTrainJobInfo(task *models.Cloudbrain) error { + if isTempJob(task.JobID, task.Status) { + if task.VersionCount > VersionCountOne { + //multi version + result, err := GetTrainJobVersionList(1000, 1, strings.TrimPrefix(task.JobID, models.TempJobIdPrefix)) + if err != nil { + log.Error("GetTrainJobVersionList failed:%v", err) + return err + } + + if result != nil { + if strconv.FormatInt(result.JobID, 10) == task.JobID && result.JobName == task.JobName { + if result.VersionCount == int64(task.VersionCount) { + log.Info("find the record(%s)", task.DisplayJobName) + task.Status = TransTrainJobStatus(result.JobVersionList[0].IntStatus) + task.VersionName = result.JobVersionList[0].VersionName + task.VersionID = result.JobVersionList[0].VersionID + + err = models.UpdateJob(task) + if err != nil { + log.Error("UpdateJob(%s) failed:%v", task.JobName, err) + return err + } + temp, err := models.GetCloudbrainTempByCloudbrainID(task.ID) + if err != nil { + log.Error("no such temp record(%s):%v", task.DisplayJobName, err.Error()) + } else { + err = models.DeleteCloudbrainTemp(temp) + if err != nil { + log.Error("DeleteCloudbrainTemp(%s) failed:%v", task.DisplayJobName, err) + } + } + + return nil + } else { + log.Error("can not find the record(%s) until now", task.DisplayJobName) + } + } else { + log.Error("can not find the record(%s) until now", task.DisplayJobName) + } + } + } else { + //inference or one version + result, err := GetTrainJobList(1000, 1, "create_time", "desc", task.JobName) + if err != nil { + log.Error("GetTrainJobList failed:%v", err) + return err + } + + if result != nil { + for _, job := range result.JobList { + if task.JobName == job.JobName { + log.Info("find the record(%s)", task.DisplayJobName) + task.Status = TransTrainJobStatus(job.IntStatus) + task.JobID = strconv.FormatInt(job.JobID, 10) + + err = models.UpdateJob(task) + if err != nil { + log.Error("UpdateJob(%s) failed:%v", task.DisplayJobName, err) + return err + } + temp, err := models.GetCloudbrainTempByCloudbrainID(task.ID) + if err != nil { + log.Error("no such temp record(%s):%v", task.DisplayJobName, err.Error()) + return err + } + err = models.DeleteCloudbrainTemp(temp) + if err != nil { + log.Error("DeleteCloudbrainTemp(%s) failed:%v", task.DisplayJobName, err) + return err + } + return nil + } + } + } + + } + } else { + //normal + result, err := GetTrainJob(task.JobID, strconv.FormatInt(task.VersionID, 10)) + if err != nil { + log.Error("GetTrainJob(%s) failed:%v", task.DisplayJobName, err) + return err + } + + if result != nil { + task.Status = TransTrainJobStatus(result.IntStatus) + task.Duration = result.Duration / 1000 + task.TrainJobDuration = result.TrainJobDuration + + if task.StartTime == 0 && result.StartTime > 0 { + task.StartTime = timeutil.TimeStamp(result.StartTime / 1000) + } + task.TrainJobDuration = models.ConvertDurationToStr(task.Duration) + if task.EndTime == 0 && models.IsTrainJobTerminal(task.Status) && task.StartTime > 0 { + task.EndTime = task.StartTime.Add(task.Duration) + } + task.CorrectCreateUnix() + err = models.UpdateJob(task) + if err != nil { + log.Error("UpdateJob(%s) failed:%v", task.JobName, err) + return err + } + } + } + + return nil +} + +func HandleNotebookInfo(task *models.Cloudbrain) error { + if isTempJob(task.JobID, task.Status) { + result, err := GetNotebookList(1000, 0, "createTime", "DESC", task.JobName) + if err != nil { + log.Error("GetNotebookList failed:%v", err) + return err + } + + if result != nil { + count, err := models.GetCloudbrainCountByJobName(task.JobName, task.JobType) + if err != nil { + log.Error("GetCloudbrainCountByJobName failed:%v", err) + return err + } + + if len(result.NotebookList) == count { + if result.NotebookList[0].JobName == task.JobName { + log.Info("find the record(%s)", task.DisplayJobName) + task.Status = result.NotebookList[0].Status + task.JobID = result.NotebookList[0].JobID + + err = models.UpdateJob(task) + if err != nil { + log.Error("UpdateJob(%s) failed:%v", task.JobName, err) + return err + } + temp, err := models.GetCloudbrainTempByCloudbrainID(task.ID) + if err != nil { + log.Error("no such temp record(%s):%v", task.DisplayJobName, err.Error()) + return err + } + err = models.DeleteCloudbrainTemp(temp) + if err != nil { + log.Error("DeleteCloudbrainTemp(%s) failed:%v", task.DisplayJobName, err) + return err + } + return nil + } else { + log.Error("can not find the record(%s) until now", task.DisplayJobName) + } + } else { + log.Error("can not find the record(%s) until now", task.DisplayJobName) + } + } else { + log.Error("can not find the record(%s) until now", task.DisplayJobName) + } + } else { + //normal + result, err := GetNotebook2(task.JobID) + if err != nil { + log.Error("GetNotebook2(%s) failed:%v", task.DisplayJobName, err) + return err + } + + if result != nil { + task.Status = result.Status + if task.StartTime == 0 && result.Lease.UpdateTime > 0 { + task.StartTime = timeutil.TimeStamp(result.Lease.UpdateTime / 1000) + } + if task.EndTime == 0 && models.IsModelArtsDebugJobTerminal(task.Status) { + task.EndTime = timeutil.TimeStampNow() + } + task.CorrectCreateUnix() + task.ComputeAndSetDuration() + err = models.UpdateJob(task) + if err != nil { + log.Error("UpdateJob(%s) failed:%v", task.DisplayJobName, err) + return err + } + } + } + + return nil +} + +func isTempJob(jobID, status string) bool { + if (strings.HasPrefix(jobID, models.TempJobIdPrefix) && status == string(models.ModelArtsTrainJobWaiting)) || status == models.JobStatusTemp { + return true + } + return false +} diff --git a/modules/modelarts/resty.go b/modules/modelarts/resty.go index 46c273a8b..babcff1aa 100755 --- a/modules/modelarts/resty.go +++ b/modules/modelarts/resty.go @@ -37,6 +37,7 @@ const ( NotebookNotFound = "ModelArts.6404" NotebookNoPermission = "ModelArts.6407" NotebookInvalid = "ModelArts.6400" + UnknownErrorPrefix = "UNKNOWN:" ) func getRestyClient() *resty.Client { @@ -298,6 +299,10 @@ sendjob: return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error()) } + if res.StatusCode() == http.StatusBadGateway { + return &result, fmt.Errorf(UnknownErrorPrefix+"createNotebook2 failed(%s): %s", response.ErrorCode, response.ErrorMsg) + } + if len(response.ErrorCode) != 0 { log.Error("ManageNotebook2 failed(%s): %s", response.ErrorCode, response.ErrorMsg) if response.ErrorCode == modelartsIllegalToken && retry < 1 { @@ -547,9 +552,6 @@ sendjob: return nil, fmt.Errorf("resty create train-job: %s", err) } - req, _ := json.Marshal(createJobParams) - log.Info("%s", req) - if res.StatusCode() == http.StatusUnauthorized && retry < 1 { retry++ _ = getToken() @@ -563,17 +565,21 @@ sendjob: return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error()) } log.Error("createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) - BootFileErrorMsg := "Invalid OBS path '" + createJobParams.Config.BootFileUrl + "'." - DataSetErrorMsg := "Invalid OBS path '" + createJobParams.Config.DataUrl + "'." - if temp.ErrorMsg == BootFileErrorMsg { + bootFileErrorMsg := "Invalid OBS path '" + createJobParams.Config.BootFileUrl + "'." + dataSetErrorMsg := "Invalid OBS path '" + createJobParams.Config.DataUrl + "'." + if temp.ErrorMsg == bootFileErrorMsg { log.Error("启动文件错误!createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) return &result, fmt.Errorf("启动文件错误!") } - if temp.ErrorMsg == DataSetErrorMsg { + if temp.ErrorMsg == dataSetErrorMsg { log.Error("数据集错误!createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) return &result, fmt.Errorf("数据集错误!") } - return &result, fmt.Errorf("createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + if res.StatusCode() == http.StatusBadGateway { + return &result, fmt.Errorf(UnknownErrorPrefix+"createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + } else { + return &result, fmt.Errorf("createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + } } if !result.IsSuccess { @@ -603,9 +609,6 @@ sendjob: return nil, fmt.Errorf("resty create train-job version: %s", err) } - req, _ := json.Marshal(createJobVersionParams) - log.Info("%s", req) - if res.StatusCode() == http.StatusUnauthorized && retry < 1 { retry++ _ = getToken() @@ -618,17 +621,23 @@ sendjob: log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error()) return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error()) } - BootFileErrorMsg := "Invalid OBS path '" + createJobVersionParams.Config.BootFileUrl + "'." - DataSetErrorMsg := "Invalid OBS path '" + createJobVersionParams.Config.DataUrl + "'." - if temp.ErrorMsg == BootFileErrorMsg { + + log.Error("createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + bootFileErrorMsg := "Invalid OBS path '" + createJobVersionParams.Config.BootFileUrl + "'." + dataSetErrorMsg := "Invalid OBS path '" + createJobVersionParams.Config.DataUrl + "'." + if temp.ErrorMsg == bootFileErrorMsg { log.Error("启动文件错误!createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) return &result, fmt.Errorf("启动文件错误!") } - if temp.ErrorMsg == DataSetErrorMsg { + if temp.ErrorMsg == dataSetErrorMsg { log.Error("数据集错误!createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) return &result, fmt.Errorf("数据集错误!") } - return &result, fmt.Errorf("createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + if res.StatusCode() == http.StatusBadGateway { + return &result, fmt.Errorf(UnknownErrorPrefix+"createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + } else { + return &result, fmt.Errorf("createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + } } if !result.IsSuccess { @@ -761,9 +770,6 @@ sendjob: goto sendjob } - //temp, _ := json.Marshal(req) - //log.Info("%s", temp) - if res.StatusCode() != http.StatusOK { var temp models.ErrorResult if err = json.Unmarshal([]byte(res.String()), &temp); err != nil { @@ -1172,7 +1178,11 @@ sendjob: log.Error("数据集错误!createInferenceJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) return &result, fmt.Errorf("数据集错误!") } - return &result, fmt.Errorf("createInferenceJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + if res.StatusCode() == http.StatusBadGateway { + return &result, fmt.Errorf(UnknownErrorPrefix+"createInferenceJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + } else { + return &result, fmt.Errorf("createInferenceJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + } } if !result.IsSuccess { @@ -1212,7 +1222,11 @@ sendjob: err = json.Unmarshal(res.Body(), &response) if err != nil { log.Error("json.Unmarshal failed: %s", err.Error()) - return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error()) + return &result, fmt.Errorf("json.Unmarshal failed: %s", err.Error()) + } + + if res.StatusCode() == http.StatusBadGateway { + return &result, fmt.Errorf(UnknownErrorPrefix+"createNotebook2 failed(%s): %s", response.ErrorCode, response.ErrorMsg) } if len(response.ErrorCode) != 0 { @@ -1271,3 +1285,139 @@ sendjob: return &result, nil } + +func GetTrainJobList(perPage, page int, sortBy, order, searchContent string) (*models.GetTrainJobListResult, error) { + checkSetting() + client := getRestyClient() + var result models.GetTrainJobListResult + + retry := 0 + +sendjob: + res, err := client.R(). + SetQueryParams(map[string]string{ + "per_page": strconv.Itoa(perPage), + "page": strconv.Itoa(page), + "sortBy": sortBy, + "order": order, + "search_content": searchContent, + }). + SetAuthToken(TOKEN). + SetResult(&result). + Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob) + + if err != nil { + return nil, fmt.Errorf("resty GetTrainJobList: %v", err) + } + + if res.StatusCode() == http.StatusUnauthorized && retry < 1 { + retry++ + _ = getToken() + goto sendjob + } + + if res.StatusCode() != http.StatusOK { + var temp models.ErrorResult + if err = json.Unmarshal([]byte(res.String()), &temp); err != nil { + log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + } + log.Error("GetTrainJobList failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + return &result, fmt.Errorf(temp.ErrorMsg) + } + + if !result.IsSuccess { + log.Error("GetTrainJobList failed(%s): %s", result.ErrorCode, result.ErrorMsg) + return &result, fmt.Errorf(result.ErrorMsg) + } + + return &result, nil +} + +func GetTrainJobVersionList(perPage, page int, jobID string) (*models.GetTrainJobVersionListResult, error) { + checkSetting() + client := getRestyClient() + var result models.GetTrainJobVersionListResult + + retry := 0 + +sendjob: + res, err := client.R(). + SetQueryParams(map[string]string{ + "per_page": strconv.Itoa(perPage), + "page": strconv.Itoa(page), + }). + SetAuthToken(TOKEN). + SetResult(&result). + Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions") + + if err != nil { + return nil, fmt.Errorf("resty GetTrainJobVersionList: %v", err) + } + + if res.StatusCode() == http.StatusUnauthorized && retry < 1 { + retry++ + _ = getToken() + goto sendjob + } + + if res.StatusCode() != http.StatusOK { + var temp models.ErrorResult + if err = json.Unmarshal([]byte(res.String()), &temp); err != nil { + log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + } + log.Error("GetTrainJobVersionList failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + return &result, fmt.Errorf(temp.ErrorMsg) + } + + if !result.IsSuccess { + log.Error("GetTrainJobVersionList failed(%s): %s", result.ErrorCode, result.ErrorMsg) + return &result, fmt.Errorf(result.ErrorMsg) + } + + return &result, nil +} + +func GetNotebookList(limit, offset int, sortBy, order, searchContent string) (*models.GetNotebookListResult, error) { + checkSetting() + client := getRestyClient() + var result models.GetNotebookListResult + + retry := 0 + +sendjob: + res, err := client.R(). + SetQueryParams(map[string]string{ + "limit": strconv.Itoa(limit), + "offset": strconv.Itoa(offset), + "name": searchContent, + "sort_key": sortBy, + "sort_dir": order, + }). + SetAuthToken(TOKEN). + SetResult(&result). + Get(HOST + "/v1/" + setting.ProjectID + urlNotebook2) + + if err != nil { + return nil, fmt.Errorf("resty GetNotebookList: %v", err) + } + + if res.StatusCode() == http.StatusUnauthorized && retry < 1 { + retry++ + _ = getToken() + goto sendjob + } + + if res.StatusCode() != http.StatusOK { + var temp models.ErrorResult + if err = json.Unmarshal([]byte(res.String()), &temp); err != nil { + log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error()) + } + log.Error("GetNotebookList failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) + return &result, fmt.Errorf(temp.ErrorMsg) + } + + return &result, nil +} diff --git a/routers/api/v1/repo/modelarts.go b/routers/api/v1/repo/modelarts.go index 54220fbca..f41d49bc1 100755 --- a/routers/api/v1/repo/modelarts.go +++ b/routers/api/v1/repo/modelarts.go @@ -26,40 +26,6 @@ import ( routerRepo "code.gitea.io/gitea/routers/repo" ) -func GetModelArtsNotebook(ctx *context.APIContext) { - var ( - err error - ) - - jobID := ctx.Params(":jobid") - repoID := ctx.Repo.Repository.ID - job, err := models.GetRepoCloudBrainByJobID(repoID, jobID) - if err != nil { - ctx.NotFound(err) - return - } - result, err := modelarts.GetJob(jobID) - if err != nil { - ctx.NotFound(err) - return - } - oldStatus := job.Status - job.Status = result.Status - if oldStatus != result.Status { - notification.NotifyChangeCloudbrainStatus(job, oldStatus) - } - err = models.UpdateJob(job) - if err != nil { - log.Error("UpdateJob failed:", err) - } - - ctx.JSON(http.StatusOK, map[string]interface{}{ - "JobID": jobID, - "JobStatus": result.Status, - }) - -} - func GetModelArtsNotebook2(ctx *context.APIContext) { var ( err error @@ -71,70 +37,16 @@ func GetModelArtsNotebook2(ctx *context.APIContext) { ctx.NotFound(err) return } - result, err := modelarts.GetNotebook2(job.JobID) + err = modelarts.HandleNotebookInfo(job) if err != nil { ctx.NotFound(err) return } - if job.StartTime == 0 && result.Lease.UpdateTime > 0 { - job.StartTime = timeutil.TimeStamp(result.Lease.UpdateTime / 1000) - } - oldStatus := job.Status - job.Status = result.Status - if job.EndTime == 0 && models.IsModelArtsDebugJobTerminal(job.Status) { - job.EndTime = timeutil.TimeStampNow() - } - job.CorrectCreateUnix() - job.ComputeAndSetDuration() - if oldStatus != result.Status { - notification.NotifyChangeCloudbrainStatus(job, oldStatus) - } - err = models.UpdateJob(job) - if err != nil { - log.Error("UpdateJob failed:", err) - } ctx.JSON(http.StatusOK, map[string]interface{}{ "ID": ID, "JobName": job.JobName, - "JobStatus": result.Status, - }) - -} - -func GetModelArtsTrainJob(ctx *context.APIContext) { - var ( - err error - ) - - jobID := ctx.Params(":jobid") - repoID := ctx.Repo.Repository.ID - job, err := models.GetRepoCloudBrainByJobID(repoID, jobID) - if err != nil { - ctx.NotFound(err) - return - } - result, err := modelarts.GetTrainJob(jobID, strconv.FormatInt(job.VersionID, 10)) - if err != nil { - ctx.NotFound(err) - return - } - oldStatus := job.Status - job.Status = modelarts.TransTrainJobStatus(result.IntStatus) - job.Duration = result.Duration - job.TrainJobDuration = result.TrainJobDuration - if oldStatus != job.Status { - notification.NotifyChangeCloudbrainStatus(job, oldStatus) - } - err = models.UpdateJob(job) - if err != nil { - log.Error("UpdateJob failed:", err) - } - - ctx.JSON(http.StatusOK, map[string]interface{}{ - "JobID": jobID, - "JobStatus": job.Status, - "JobDuration": job.Duration, + "JobStatus": job.Status, }) } @@ -188,27 +100,11 @@ func GetModelArtsTrainJobVersion(ctx *context.APIContext) { } } } else if job.Type == models.TypeCloudBrainTwo { - result, err := modelarts.GetTrainJob(jobID, strconv.FormatInt(job.VersionID, 10)) + err := modelarts.HandleTrainJobInfo(job) if err != nil { ctx.NotFound(err) return } - - if job.StartTime == 0 && result.StartTime > 0 { - job.StartTime = timeutil.TimeStamp(result.StartTime / 1000) - } - job.Status = modelarts.TransTrainJobStatus(result.IntStatus) - job.Duration = result.Duration / 1000 - job.TrainJobDuration = models.ConvertDurationToStr(job.Duration) - - if job.EndTime == 0 && models.IsTrainJobTerminal(job.Status) && job.StartTime > 0 { - job.EndTime = job.StartTime.Add(job.Duration) - } - job.CorrectCreateUnix() - err = models.UpdateTrainJobVersion(job) - if err != nil { - log.Error("UpdateJob failed:", err) - } } else if job.Type == models.TypeC2Net { result, err := grampus.GetJob(jobID) if err != nil { @@ -557,26 +453,11 @@ func GetModelArtsInferenceJob(ctx *context.APIContext) { ctx.NotFound(err) return } - result, err := modelarts.GetTrainJob(jobID, strconv.FormatInt(job.VersionID, 10)) + err = modelarts.HandleTrainJobInfo(job) if err != nil { ctx.NotFound(err) return } - if job.StartTime == 0 && result.StartTime > 0 { - job.StartTime = timeutil.TimeStamp(result.StartTime / 1000) - } - job.Status = modelarts.TransTrainJobStatus(result.IntStatus) - job.Duration = result.Duration / 1000 - job.TrainJobDuration = models.ConvertDurationToStr(job.Duration) - - if job.EndTime == 0 && models.IsTrainJobTerminal(job.Status) && job.StartTime > 0 { - job.EndTime = job.StartTime.Add(job.Duration) - } - job.CorrectCreateUnix() - err = models.UpdateInferenceJob(job) - if err != nil { - log.Error("UpdateJob failed:", err) - } ctx.JSON(http.StatusOK, map[string]interface{}{ "JobID": jobID, diff --git a/routers/repo/cloudbrain.go b/routers/repo/cloudbrain.go index da9801664..79d561ce7 100755 --- a/routers/repo/cloudbrain.go +++ b/routers/repo/cloudbrain.go @@ -1784,70 +1784,24 @@ func SyncCloudbrainStatus() { } } else if task.Type == models.TypeCloudBrainTwo { if task.JobType == string(models.JobTypeDebug) { - //result, err := modelarts.GetJob(task.JobID) - result, err := modelarts.GetNotebook2(task.JobID) + err := modelarts.HandleNotebookInfo(task) if err != nil { - log.Error("GetJob(%s) failed:%v", task.JobName, err) + log.Error("HandleNotebookInfo(%s) failed:%v", task.DisplayJobName, err) continue } - - if result != nil { - oldStatus := task.Status - task.Status = result.Status - if task.StartTime == 0 && result.Lease.UpdateTime > 0 { - task.StartTime = timeutil.TimeStamp(result.Lease.UpdateTime / 1000) - } - if task.EndTime == 0 && models.IsModelArtsDebugJobTerminal(task.Status) { - task.EndTime = timeutil.TimeStampNow() - } - task.CorrectCreateUnix() - task.ComputeAndSetDuration() - if oldStatus != task.Status { - notification.NotifyChangeCloudbrainStatus(task, oldStatus) - } - err = models.UpdateJob(task) - if err != nil { - log.Error("UpdateJob(%s) failed:%v", task.JobName, err) - continue - } - } } else if task.JobType == string(models.JobTypeTrain) || task.JobType == string(models.JobTypeInference) { - result, err := modelarts.GetTrainJob(task.JobID, strconv.FormatInt(task.VersionID, 10)) + err := modelarts.HandleTrainJobInfo(task) if err != nil { - log.Error("GetTrainJob(%s) failed:%v", task.JobName, err) + log.Error("HandleTrainJobInfo(%s) failed:%v", task.DisplayJobName, err) continue } - - if result != nil { - oldStatus := task.Status - task.Status = modelarts.TransTrainJobStatus(result.IntStatus) - task.Duration = result.Duration / 1000 - task.TrainJobDuration = result.TrainJobDuration - - if task.StartTime == 0 && result.StartTime > 0 { - task.StartTime = timeutil.TimeStamp(result.StartTime / 1000) - } - task.TrainJobDuration = models.ConvertDurationToStr(task.Duration) - if task.EndTime == 0 && models.IsTrainJobTerminal(task.Status) && task.StartTime > 0 { - task.EndTime = task.StartTime.Add(task.Duration) - } - task.CorrectCreateUnix() - if oldStatus != task.Status { - notification.NotifyChangeCloudbrainStatus(task, oldStatus) - } - 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) + log.Error("task.JobType(%s) is error:%s", task.DisplayJobName, task.JobType) } } else if task.Type == models.TypeC2Net { result, err := grampus.GetJob(task.JobID) if err != nil { - log.Error("GetTrainJob(%s) failed:%v", task.JobName, err) + log.Error("GetTrainJob(%s) failed:%v", task.DisplayJobName, err) continue } diff --git a/routers/repo/grampus.go b/routers/repo/grampus.go index 53fd9be30..463cc56e3 100755 --- a/routers/repo/grampus.go +++ b/routers/repo/grampus.go @@ -337,7 +337,7 @@ func GrampusTrainJobGpuCreate(ctx *context.Context, form auth.CreateGrampusTrain EngineName: image, DatasetName: attachment.Name, IsLatestVersion: modelarts.IsLatestVersion, - VersionCount: modelarts.VersionCount, + VersionCount: modelarts.VersionCountOne, WorkServerNumber: 1, } @@ -387,7 +387,7 @@ func GrampusTrainJobNpuCreate(ctx *context.Context, form auth.CreateGrampusTrain branchName := form.BranchName isLatestVersion := modelarts.IsLatestVersion flavorName := form.FlavorName - versionCount := modelarts.VersionCount + versionCount := modelarts.VersionCountOne engineName := form.EngineName if !jobNamePattern.MatchString(displayJobName) { diff --git a/routers/repo/modelarts.go b/routers/repo/modelarts.go index 88bf0d22a..f902244ea 100755 --- a/routers/repo/modelarts.go +++ b/routers/repo/modelarts.go @@ -7,6 +7,7 @@ import ( "fmt" "io" "io/ioutil" + "math/rand" "net/http" "os" "path" @@ -262,30 +263,15 @@ func NotebookShow(ctx *context.Context) { return } - result, err := modelarts.GetNotebook2(task.JobID) - if err != nil { - log.Error("GET job error", err.Error()) - ctx.NotFound(ctx.Req.URL.RequestURI(), nil) - return - } - - if result != nil { - if task.DeletedAt.IsZero() { //normal record - if task.Status != result.Status { - oldStatus := task.Status - task.Status = result.Status - models.ParseAndSetDurationFromModelArtsNotebook(result, task) - notification.NotifyChangeCloudbrainStatus(task, oldStatus) - err = models.UpdateJob(task) - if err != nil { - log.Error("GET job error", err.Error()) - ctx.NotFound(ctx.Req.URL.RequestURI(), nil) - return - } - } - } else { //deleted record - + if task.DeletedAt.IsZero() { //normal record + err := modelarts.HandleNotebookInfo(task) + if err != nil { + ctx.Data["error"] = err.Error() + ctx.RenderWithErr(err.Error(), tplModelArtsNotebookShow, nil) + return } + } else { //deleted record + } datasetDownload := make([]models.DatasetDownload, 0) @@ -396,82 +382,141 @@ func NotebookDebug2(ctx *context.Context) { ctx.Redirect(result.Url + "?token=" + result.Token) } -func NotebookManage(ctx *context.Context) { +func NotebookRestart(ctx *context.Context) { var ID = ctx.Params(":id") - var action = ctx.Params(":action") - var resultCode = "0" + var resultCode = "-1" var errorMsg = "" var status = "" + task := ctx.Cloudbrain + for { - task, err := models.GetCloudbrainByID(ID) - if err != nil { - log.Error("get task(%s) failed:%v", task.JobName, err.Error(), ctx.Data["MsgID"]) - resultCode = "-1" - errorMsg = "system error" + ctx.CheckWechatBind() + if ctx.Written() { + return + } + if task.Status != string(models.ModelArtsStopped) && task.Status != string(models.ModelArtsStartFailed) && task.Status != string(models.ModelArtsCreateFailed) { + log.Error("the job(%s) is not stopped", task.JobName, ctx.Data["MsgID"]) + errorMsg = "the job is not stopped" break } - if action == models.ActionStop { - if task.Status != string(models.ModelArtsRunning) { - log.Error("the job(%s) is not running", task.JobName, ctx.Data["MsgID"]) - resultCode = "-1" - errorMsg = "the job is not running" + count, err := models.GetCloudbrainNotebookCountByUserID(ctx.User.ID) + if err != nil { + log.Error("GetCloudbrainNotebookCountByUserID failed:%v", err, ctx.Data["MsgID"]) + errorMsg = "system error" + break + } else { + if count >= 1 { + log.Error("the user already has running or waiting task", ctx.Data["MsgID"]) + errorMsg = "you have already a running or waiting task, can not create more" break } + } - if !ctx.IsSigned || (ctx.User.ID != task.UserID && !ctx.IsUserSiteAdmin() && !ctx.IsUserRepoOwner()) { - log.Error("the user has no right ro stop the job", task.JobName, ctx.Data["MsgID"]) - resultCode = "-1" - errorMsg = "you have no right to stop the job" - break - } - } else if action == models.ActionRestart { - ctx.CheckWechatBind() - if ctx.Written() { - return - } - if task.Status != string(models.ModelArtsStopped) && task.Status != string(models.ModelArtsStartFailed) && task.Status != string(models.ModelArtsCreateFailed) { - log.Error("the job(%s) is not stopped", task.JobName, ctx.Data["MsgID"]) - resultCode = "-1" - errorMsg = "the job is not stopped" - break - } + createTime := timeutil.TimeStampNow() + newTask := &models.Cloudbrain{ + Status: string(models.ModelArtsTrainJobWaiting), + UserID: task.UserID, + RepoID: task.RepoID, + JobID: models.TempJobIdPrefix + task.JobName + strconv.Itoa(int(rand.New(rand.NewSource(time.Now().UnixNano())).Int31n(100000))), + JobName: task.JobName, + DisplayJobName: task.DisplayJobName, + JobType: task.JobType, + Type: task.Type, + Uuid: task.Uuid, + Image: task.Image, + ComputeResource: task.ComputeResource, + Description: task.Description, + CreatedUnix: createTime, + UpdatedUnix: createTime, + } - if !ctx.IsSigned || (ctx.User.ID != task.UserID && !ctx.IsUserSiteAdmin()) { - log.Error("the user has no right ro restart the job", task.JobName, ctx.Data["MsgID"]) - resultCode = "-1" - errorMsg = "you have no right to restart the job" + err = models.RestartCloudbrain(task, newTask) + if err != nil { + log.Error("RestartCloudbrain(%s) failed:%v", task.JobName, err.Error(), ctx.Data["MsgID"]) + errorMsg = "system error" + break + } + + param := models.NotebookAction{ + Action: models.ActionStart, + } + + res, err := modelarts.ManageNotebook2(task.JobID, param) + if err != nil { + log.Error("ManageNotebook2(%s) failed:%v", task.DisplayJobName, err.Error(), ctx.Data["MsgID"]) + if strings.HasPrefix(err.Error(), modelarts.UnknownErrorPrefix) { + log.Info("(%s)unknown error, set temp status", newTask.DisplayJobName) + errTemp := models.InsertCloudbrainTemp(&models.CloudbrainTemp{ + CloudbrainID: newTask.ID, + Status: models.JobStatusTemp, + Type: newTask.Type, + JobName: newTask.JobName, + JobType: newTask.JobType, + }) + if errTemp != nil { + log.Error("InsertCloudbrainTemp failed: %v", errTemp.Error()) + } + } else { + newTask.Status = string(models.ModelArtsTrainJobFailed) + errTemp := models.UpdateJob(newTask) + if errTemp != nil { + log.Error("UpdateJob failed: %v", errTemp.Error()) + } + errTemp = models.DeleteJob(newTask) + if errTemp != nil { + log.Error("DeleteJob failed: %v", errTemp.Error()) + } + errorMsg = err.Error() break } - - count, err := models.GetCloudbrainNotebookCountByUserID(ctx.User.ID) + } else { + newTask.Status = res.Status + newTask.JobID = task.JobID + err = models.UpdateJob(newTask) if err != nil { - log.Error("GetCloudbrainNotebookCountByUserID failed:%v", err, ctx.Data["MsgID"]) - resultCode = "-1" - errorMsg = "system error" + log.Error("UpdateJob failed: %v", err.Error()) + errorMsg = err.Error() break - } else { - if count >= 1 { - log.Error("the user already has running or waiting task", ctx.Data["MsgID"]) - resultCode = "-1" - errorMsg = "you have already a running or waiting task, can not create more" - break - } } + } - action = models.ActionStart - } else { - log.Error("the action(%s) is illegal", action, ctx.Data["MsgID"]) + status = res.Status + resultCode = "0" + notification.NotifyOtherTask(ctx.User, ctx.Repo.Repository, strconv.FormatInt(newTask.ID, 10), newTask.DisplayJobName, models.ActionCreateDebugNPUTask) + + break + } + + ctx.JSON(200, map[string]string{ + "result_code": resultCode, + "error_msg": errorMsg, + "status": status, + "id": ID, + }) +} + +func NotebookStop(ctx *context.Context) { + var ID = ctx.Params(":id") + var resultCode = "0" + var errorMsg = "" + var status = "" + + task := ctx.Cloudbrain + + for { + if task.Status != string(models.ModelArtsRunning) { + log.Error("the job(%s) is not running", task.JobName, ctx.Data["MsgID"]) resultCode = "-1" - errorMsg = "非法操作" + errorMsg = "the job is not running" break } param := models.NotebookAction{ - Action: action, + Action: models.ActionStop, } - createTime := timeutil.TimeStampNow() + res, err := modelarts.ManageNotebook2(task.JobID, param) if err != nil { log.Error("ManageNotebook2(%s) failed:%v", task.JobName, err.Error(), ctx.Data["MsgID"]) @@ -484,50 +529,17 @@ func NotebookManage(ctx *context.Context) { } status = res.Status - if action == models.ActionStart { - newTask := &models.Cloudbrain{ - Status: status, - UserID: task.UserID, - RepoID: task.RepoID, - JobID: task.JobID, - JobName: task.JobName, - DisplayJobName: task.DisplayJobName, - JobType: task.JobType, - Type: task.Type, - Uuid: task.Uuid, - Image: task.Image, - ComputeResource: task.ComputeResource, - Description: task.Description, - CreatedUnix: createTime, - UpdatedUnix: createTime, - } - - err = models.RestartCloudbrain(task, newTask) - if err != nil { - log.Error("RestartCloudbrain(%s) failed:%v", task.JobName, err.Error(), ctx.Data["MsgID"]) - resultCode = "-1" - errorMsg = "system error" - break - } - ID = strconv.FormatInt(newTask.ID, 10) - notification.NotifyOtherTask(ctx.User, ctx.Repo.Repository, ID, task.DisplayJobName, models.ActionCreateDebugNPUTask) - } else { - oldStatus := task.Status - task.Status = res.Status - if task.EndTime == 0 && models.IsModelArtsDebugJobTerminal(task.Status) { - task.EndTime = timeutil.TimeStampNow() - } - task.ComputeAndSetDuration() - if oldStatus != task.Status { - notification.NotifyChangeCloudbrainStatus(task, oldStatus) - } - err = models.UpdateJob(task) - if err != nil { - log.Error("UpdateJob(%s) failed:%v", task.JobName, err.Error(), ctx.Data["MsgID"]) - resultCode = "-1" - errorMsg = "system error" - break - } + task.Status = res.Status + if task.EndTime == 0 && models.IsModelArtsDebugJobTerminal(task.Status) { + task.EndTime = timeutil.TimeStampNow() + } + task.ComputeAndSetDuration() + err = models.UpdateJob(task) + if err != nil { + log.Error("UpdateJob(%s) failed:%v", task.JobName, err.Error(), ctx.Data["MsgID"]) + resultCode = "-1" + errorMsg = "system error" + break } break @@ -1000,7 +1012,7 @@ func TrainJobCreate(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) branch_name := form.BranchName isLatestVersion := modelarts.IsLatestVersion FlavorName := form.FlavorName - VersionCount := modelarts.VersionCount + VersionCount := modelarts.VersionCountOne EngineName := form.EngineName count, err := models.GetCloudbrainTrainJobCountByUserID(ctx.User.ID) @@ -1702,60 +1714,6 @@ func TrainJobShow(ctx *context.Context) { ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow) } -func TrainJobGetLog(ctx *context.Context) { - ctx.Data["PageIsTrainJob"] = true - - var jobID = ctx.Params(":jobid") - var logFileName = ctx.Query("file_name") - var baseLine = ctx.Query("base_line") - var order = ctx.Query("order") - - if order != modelarts.OrderDesc && order != modelarts.OrderAsc { - log.Error("order(%s) check failed", order) - ctx.HTML(http.StatusBadRequest, tplModelArtsTrainJobShow) - return - } - - task, err := models.GetCloudbrainByJobID(jobID) - if err != nil { - log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error()) - ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil) - return - } - - result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), baseLine, logFileName, order, modelarts.Lines) - if err != nil { - log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error()) - ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil) - return - } - - ctx.Data["log"] = result - //ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow) -} - -func trainJobGetLog(jobID string) (*models.GetTrainJobLogFileNamesResult, *models.GetTrainJobLogResult, error) { - task, err := models.GetCloudbrainByJobID(jobID) - if err != nil { - log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error()) - return nil, nil, err - } - - resultLogFile, err := modelarts.GetTrainJobLogFileNames(jobID, strconv.FormatInt(task.VersionID, 10)) - if err != nil { - log.Error("GetTrainJobLogFileNames(%s) failed:%v", jobID, err.Error()) - return nil, nil, err - } - - result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), "", resultLogFile.LogFileList[0], modelarts.OrderDesc, modelarts.Lines) - if err != nil { - log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error()) - return nil, nil, err - } - - return resultLogFile, result, err -} - func TrainJobDel(ctx *context.Context) { var jobID = ctx.Params(":jobid") var listType = ctx.Query("listType") @@ -1822,15 +1780,6 @@ func TrainJobStop(ctx *context.Context) { ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job?listType=" + listType) } -func canUserCreateTrainJob(uid int64) (bool, error) { - org, err := models.GetOrgByName(setting.AllowedOrg) - if err != nil { - log.Error("get allowed org failed: ", setting.AllowedOrg) - return false, err - } - - return org.IsOrgMember(uid) -} func canUserCreateTrainJobVersion(ctx *context.Context, userID int64) (bool, error) { if ctx == nil || ctx.User == nil { log.Error("user unlogin!") @@ -1922,7 +1871,7 @@ func InferenceJobCreate(ctx *context.Context, form auth.CreateModelArtsInference EngineName := form.EngineName LabelName := form.LabelName isLatestVersion := modelarts.IsLatestVersion - VersionCount := modelarts.VersionCount + VersionCount := modelarts.VersionCountOne trainUrl := form.TrainUrl modelName := form.ModelName modelVersion := form.ModelVersion diff --git a/routers/routes/routes.go b/routers/routes/routes.go index f917aebf1..50443b1c3 100755 --- a/routers/routes/routes.go +++ b/routers/routes/routes.go @@ -1183,7 +1183,8 @@ func RegisterRoutes(m *macaron.Macaron) { m.Group("/:id", func() { m.Get("", reqRepoCloudBrainReader, repo.NotebookShow) m.Get("/debug", cloudbrain.AdminOrJobCreaterRight, repo.NotebookDebug2) - m.Post("/:action", reqRepoCloudBrainWriter, repo.NotebookManage) + m.Post("/restart", cloudbrain.AdminOrJobCreaterRight, repo.NotebookRestart) + m.Post("/stop", cloudbrain.AdminOrJobCreaterRight, repo.NotebookStop) m.Post("/del", cloudbrain.AdminOrOwnerOrJobCreaterRight, repo.NotebookDel) }) m.Get("/create", reqWechatBind, reqRepoCloudBrainWriter, repo.NotebookNew)