diff --git a/models/attachment.go b/models/attachment.go index 75e937913..d217a61a4 100755 --- a/models/attachment.go +++ b/models/attachment.go @@ -379,7 +379,7 @@ func GetUnDecompressAttachments() ([]*Attachment, error) { func getUnDecompressAttachments(e Engine) ([]*Attachment, error) { attachments := make([]*Attachment, 0, 10) - return attachments, e.Where("decompress_state = ? and dataset_id != 0 and attachment.type = ? and (name like '%.zip' or name like '%.tar.gz' or name like '%.tgz')", DecompressStateInit, TypeCloudBrainOne).Find(&attachments) + return attachments, e.Where("decompress_state = ? and dataset_id != 0 and (name like '%.zip' or name like '%.tar.gz' or name like '%.tgz')", DecompressStateInit).Find(&attachments) } func GetAllPublicAttachments() ([]*AttachmentUsername, error) { @@ -429,7 +429,7 @@ func GetAllUserAttachments(userID int64) ([]*AttachmentUsername, error) { func getModelArtsUserAttachments(e Engine, userID int64) ([]*AttachmentUsername, error) { attachments := make([]*AttachmentUsername, 0, 10) if err := e.Table("attachment").Join("LEFT", "`user`", "attachment.uploader_id "+ - "= `user`.id").Where("attachment.type = ? and (uploader_id= ? or is_private = ?)", TypeCloudBrainNotebook, userID, false).Find(&attachments); err != nil { + "= `user`.id").Where("attachment.type = ? and (uploader_id= ? or is_private = ?)", TypeCloudBrainTwo, userID, false).Find(&attachments); err != nil { return nil, err } return attachments, nil diff --git a/models/cloudbrain.go b/models/cloudbrain.go index be14c633e..563ab9d06 100755 --- a/models/cloudbrain.go +++ b/models/cloudbrain.go @@ -70,6 +70,7 @@ type Cloudbrain struct { VersionID int64 `xorm:"INDEX DEFAULT 0"` VersionName string Uuid string + DatasetName string User *User `xorm:"-"` Repo *Repository `xorm:"-"` diff --git a/modules/modelarts/modelarts.go b/modules/modelarts/modelarts.go index a0d068b56..5717d044f 100755 --- a/modules/modelarts/modelarts.go +++ b/modules/modelarts/modelarts.go @@ -195,6 +195,12 @@ func GenerateTrainJob(ctx *context.Context, req *GenerateTrainJobReq) 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, @@ -206,6 +212,7 @@ func GenerateTrainJob(ctx *context.Context, req *GenerateTrainJobReq) error { VersionID: jobResult.VersionID, VersionName: jobResult.VersionName, Uuid: req.Uuid, + DatasetName: attach.Name, }) if err != nil { diff --git a/modules/modelarts/resty.go b/modules/modelarts/resty.go index ad95dd6f8..d17478c94 100755 --- a/modules/modelarts/resty.go +++ b/modules/modelarts/resty.go @@ -366,14 +366,6 @@ 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) - if temp.ErrorCode == "ModelArts.0404" { - temp.ErrorCode = "0404" - temp.ErrorMsg = "启动文件未找到！" - } - if temp.ErrorCode == "ModelArts.0104" { - temp.ErrorCode = "0104" - temp.ErrorMsg = "运行参数错误！" - } return &result, fmt.Errorf("createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg) } diff --git a/options/locale/locale_en-US.ini b/options/locale/locale_en-US.ini index 7d1d7a3e0..7fa8c7a4f 100644 --- a/options/locale/locale_en-US.ini +++ b/options/locale/locale_en-US.ini @@ -764,15 +764,15 @@ submit_image=Submit Image download=Download -cloudbrain=cloudbrain +cloudbrain=Cloudbrain cloudbrain.new=New cloudbrain -cloudbrain.desc=cloudbrain +cloudbrain.desc=Cloudbrain cloudbrain.cancel=Cancel -cloudbrain.commit_image = submit -clone_cnt=download -balance = balance -balance.total_view = total balance -balance.available = available balance： +cloudbrain.commit_image = Submit +clone_cnt=Download +balance = Balance +balance.total_view = Total Balance +balance.available = Available Balance： cloudbrain1 = cloudbrain1 cloudbrain2 = cloudbrain2 cloudbrain_selection = select cloudbrain @@ -787,7 +787,7 @@ cloudbrain_status_runtime = Running Time modelarts.notebook=Debug Task -modelarts.train_job=Create Task +modelarts.train_job=Train Task modelarts.train_job.new_debug= New Debug Task modelarts.train_job.new_train=New Train Task modelarts.train_job.config=Configuration information diff --git a/routers/repo/attachment.go b/routers/repo/attachment.go index b6661433e..5bfc4f9a2 100755 --- a/routers/repo/attachment.go +++ b/routers/repo/attachment.go @@ -917,16 +917,22 @@ func HandleUnDecompressAttachment() { } for _, attach := range attachs { - err = worker.SendDecompressTask(contexExt.Background(), attach.UUID, attach.Name) - if err != nil { - log.Error("SendDecompressTask(%s) failed:%s", attach.UUID, err.Error()) - } else { - attach.DecompressState = models.DecompressStateIng - err = models.UpdateAttachment(attach) + if attach.Type == models.TypeCloudBrainOne { + err = worker.SendDecompressTask(contexExt.Background(), attach.UUID, attach.Name) if err != nil { - log.Error("UpdateAttachment state(%s) failed:%s", attach.UUID, err.Error()) + log.Error("SendDecompressTask(%s) failed:%s", attach.UUID, err.Error()) + } else { + attach.DecompressState = models.DecompressStateIng + err = models.UpdateAttachment(attach) + if err != nil { + log.Error("UpdateAttachment state(%s) failed:%s", attach.UUID, err.Error()) + } } + } else if attach.Type == models.TypeCloudBrainTwo { + attachjson, _ := json.Marshal(attach) + labelmsg.SendDecompressAttachToLabelOBS(string(attachjson)) } + } return diff --git a/routers/repo/modelarts.go b/routers/repo/modelarts.go index b37af704d..ea8ab0c06 100755 --- a/routers/repo/modelarts.go +++ b/routers/repo/modelarts.go @@ -4,6 +4,7 @@ import ( "encoding/json" "errors" "io" + "io/ioutil" "net/http" "os" "path" @@ -35,10 +36,10 @@ const ( tplModelArtsNew base.TplName = "repo/modelarts/new" tplModelArtsShow base.TplName = "repo/modelarts/show" - tplModelArtsTrainJobIndex base.TplName = "repo/modelarts/trainjob/index" - tplModelArtsTrainJobNew base.TplName = "repo/modelarts/trainjob/new" - tplModelArtsTrainJobShow base.TplName = "repo/modelarts/trainjob/show" - tplModelArtsTrainJobShowModels base.TplName = "repo/modelarts/trainjob/models/index" + tplModelArtsTrainJobIndex base.TplName = "repo/modelarts/trainjob/index" + tplModelArtsTrainJobNew base.TplName = "repo/modelarts/trainjob/new" + tplModelArtsTrainJobShow base.TplName = "repo/modelarts/trainjob/show" + tplModelArtsTrainJobShowModels base.TplName = "repo/modelarts/trainjob/models/index" ) // MustEnableDataset check if repository enable internal cb @@ -621,7 +622,7 @@ func TrainJobCreate(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) codeObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.CodePath outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath logObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.LogPath - dataPath := "/" + setting.Bucket + "/" + setting.BasePath + path.Join(uuid[0:1], uuid[1:2]) + "/" + uuid + "/" + dataPath := "/" + setting.Bucket + "/" + setting.BasePath + path.Join(uuid[0:1], uuid[1:2]) + "/" + uuid + uuid + "/" //can, err := canUserCreateTrainJob(ctx.User.ID) //if err != nil { @@ -643,10 +644,29 @@ func TrainJobCreate(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) return } + attach, err := models.GetAttachmentByUUID(uuid) + if err != nil { + log.Error("GetAttachmentByUUID(%s) failed:%v", uuid, err.Error()) + return + } + + //todo: del the codeLocalPath + _, err = ioutil.ReadDir(codeLocalPath) + if err == nil { + os.RemoveAll(codeLocalPath) + } if err := git.Clone(repo.RepoPath(), codeLocalPath, git.CloneRepoOptions{}); err != nil { - log.Error("Failed to clone repository: %s (%v)", repo.FullName(), err) + log.Error("创建任务失败，任务名称已存在！: %s (%v)", repo.FullName(), err) trainJobNewDataPrepare(ctx) - ctx.RenderWithErr("Failed to clone repository", tplModelArtsTrainJobNew, &form) + + ctx.Data["bootFile"] = form.BootFile + ctx.Data["uuid"] = form.Attachment + ctx.Data["datasetName"] = attach.Name + ctx.Data["params"] = form.Params + trainJobNewDataPrepare(ctx) + // ctx.RenderWithErr("Failed to clone repository", tplModelArtsTrainJobNew, &form) + ctx.RenderWithErr("创建任务失败，任务名称已存在！", tplModelArtsTrainJobNew, &form) + // ctx.RenderWithErr(err, tplModelArtsTrainJobNew, &form) return } @@ -752,10 +772,14 @@ func TrainJobCreate(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) Parameters: param, } - err := modelarts.GenerateTrainJob(ctx, req) + err = modelarts.GenerateTrainJob(ctx, req) if err != nil { log.Error("GenerateTrainJob failed:%v", err.Error()) trainJobNewDataPrepare(ctx) + ctx.Data["bootFile"] = form.BootFile + ctx.Data["uuid"] = form.Attachment + ctx.Data["datasetName"] = attach.Name + ctx.Data["params"] = form.Params ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobNew, &form) return } @@ -860,12 +884,12 @@ func TrainJobShow(ctx *context.Context) { return } - attach, err := models.GetAttachmentByUUID(task.Uuid) - if err != nil { - log.Error("GetAttachmentByUUID(%s) failed:%v", jobID, err.Error()) - ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil) - return - } + // attach, err := models.GetAttachmentByUUID(task.Uuid) + // if err != nil { + // log.Error("GetAttachmentByUUID(%s) failed:%v", jobID, err.Error()) + // ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil) + // return + // } result, err := modelarts.GetTrainJob(jobID, strconv.FormatInt(task.VersionID, 10)) if err != nil { @@ -889,7 +913,7 @@ func TrainJobShow(ctx *context.Context) { return } - result.DatasetName = attach.Name + result.DatasetName = task.DatasetName } resultLogFile, resultLog, err := trainJobGetLog(jobID) diff --git a/templates/repo/modelarts/trainjob/index.tmpl b/templates/repo/modelarts/trainjob/index.tmpl index 4337d6de8..7bc14d6a4 100755 --- a/templates/repo/modelarts/trainjob/index.tmpl +++ b/templates/repo/modelarts/trainjob/index.tmpl @@ -355,7 +355,7 @@
{{$.CsrfTokenHtml}} {{if $.Permission.CanWrite $.UnitTypeCloudBrain}} - + {{$.i18n.Tr "repo.stop"}} {{else}} @@ -535,7 +535,7 @@ $('#model-delete-'+jobID).removeClass('red') $('#model-delete-'+jobID).addClass('disabled') } - if(status=="KILLED" || status=="FAILED"){ + if(status=="KILLED" || status=="FAILED" || status=="KILLING"){ $('#stop-model-debug-'+jobID).removeClass('blue') $('#stop-model-debug-'+jobID).addClass('disabled') $('#model-delete-'+jobID).removeClass('disabled') diff --git a/templates/repo/modelarts/trainjob/new.tmpl b/templates/repo/modelarts/trainjob/new.tmpl index 6e98158d9..641f6d51c 100755 --- a/templates/repo/modelarts/trainjob/new.tmpl +++ b/templates/repo/modelarts/trainjob/new.tmpl @@ -156,16 +156,12 @@

{{.i18n.Tr "repo.modelarts.train_job.basic_info"}}:

- +
- +
- +
- + {{if .bootFile}} + + {{else}} + + {{end}}
- + {{if $.uuid}} + + {{end}} {{range .attachments}} + {{end}}
- +
@@ -263,7 +267,7 @@