Merge pull request 'NPU训练任务增加日志下载功能' (#2253) from download-npu-log into V20220616

Reviewed-on: https://git.openi.org.cn/OpenI/aiforge/pulls/2253
Reviewed-by: liuzx <liuzx@pcl.ac.cn>

modules/storage/obs.go

@@ -564,3 +564,17 @@ func ObsCreateObject(path string) error {
 
 	return nil
 }
+
+func GetObsLogFileName(prefix string) (string, error) {
+	input := &obs.ListObjectsInput{}
+	input.Bucket = setting.Bucket
+	input.Prefix = prefix
+
+	output, err := ObsCli.ListObjects(input)
+	if err != nil {
+		log.Error("PutObject failed:", err.Error())
+		return "", err
+	}
+
+	return output.Contents[0].Key, nil
+}

options/locale/locale_en-US.ini

@@ -1145,6 +1145,7 @@ modelarts.infer_job.model_version = Model/Version
 modelarts.infer_job.select_model = Select Model
 modelarts.infer_job.boot_file_helper=The startup file is the entry file for your program execution and must end in.py.Such as  inference.py, main.py, example/inference. Py, case/main.py.
 modelarts.infer_job.tooltip = The model has been deleted and cannot be viewed.
+modelarts.download_log=Download log file
 
 
 debug_task_not_created = Debug task has not been created

options/locale/locale_zh-CN.ini

@@ -1155,6 +1155,7 @@ modelarts.infer_job.model_version = 模型/版本
 modelarts.infer_job.select_model = 选择模型
 modelarts.infer_job.boot_file_helper=启动文件是您程序执行的入口文件，必须是以.py结尾的文件。比如inference.py、main.py、example/inference.py、case/main.py。
 modelarts.infer_job.tooltip = 该模型已删除，无法查看。
+modelarts.download_log=下载日志文件
 
 
 debug_task_not_created = 未创建过调试任务

routers/repo/modelarts.go

@@ -2251,7 +2251,6 @@ func ModelDownload(ctx *context.Context) {
 	versionName := ctx.Query("version_name")
 	parentDir := ctx.Query("parent_dir")
 	fileName := ctx.Query("file_name")
-	log.Info("DownloadSingleModelFile start.")
 	task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
 	if err != nil {
 		log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err.Error())
@@ -2259,7 +2258,6 @@ func ModelDownload(ctx *context.Context) {
 	}
 
 	path := strings.TrimPrefix(path.Join(setting.TrainJobModelPath, task.JobName, setting.OutPutPath, versionName, parentDir, fileName), "/")
-	log.Info("Download path is:%s", path)
 
 	url, err := storage.GetObsCreateSignedUrlByBucketAndKey(setting.Bucket, path)
 	if err != nil {
@@ -2267,6 +2265,7 @@ func ModelDownload(ctx *context.Context) {
 		ctx.ServerError("GetObsCreateSignedUrl", err)
 		return
 	}
+	ctx.Resp.Header().Set("Cache-Control", "max-age=0")
 	http.Redirect(ctx.Resp, ctx.Req.Request, url, http.StatusMovedPermanently)
 }
 
@@ -2278,13 +2277,11 @@ func ResultDownload(ctx *context.Context) {
 	versionName := ctx.Query("version_name")
 	parentDir := ctx.Query("parent_dir")
 	fileName := ctx.Query("file_name")
-	log.Info("DownloadResult start.")
 	task := ctx.Cloudbrain
 	if err != nil {
 		ctx.Data["error"] = err.Error()
 	}
 	path := strings.TrimPrefix(path.Join(setting.TrainJobModelPath, task.JobName, "result/", versionName, parentDir, fileName), "/")
-	log.Info("Download path is:%s", path)
 
 	url, err := storage.GetObsCreateSignedUrlByBucketAndKey(setting.Bucket, path)
 	if err != nil {
@@ -2292,6 +2289,7 @@ func ResultDownload(ctx *context.Context) {
 		ctx.ServerError("GetObsCreateSignedUrl", err)
 		return
 	}
+	ctx.Resp.Header().Set("Cache-Control", "max-age=0")
 	http.Redirect(ctx.Resp, ctx.Req.Request, url, http.StatusMovedPermanently)
 }
 func DeleteJobStorage(jobName string) error {
@@ -2390,3 +2388,35 @@ func SetJobCount(ctx *context.Context) {
 	}
 	ctx.Data["jobCount"] = jobCount
 }
+
+func TrainJobDownloadLogFile(ctx *context.Context) {
+	var (
+		err error
+	)
+
+	var jobID = ctx.Params(":jobid")
+	versionName := ctx.Query("version_name")
+	task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
+	if err != nil {
+		log.Error("GetCloudbrainByJobIDAndVersionName(%s) failed:%v", task.JobName, err.Error(), ctx.Data["msgID"])
+		ctx.ServerError("GetCloudbrainByJobIDAndVersionName", err)
+		return
+	}
+
+	prefix := strings.TrimPrefix(path.Join(setting.TrainJobModelPath, task.JobName, modelarts.LogPath, versionName), "/") + "/job"
+	key, err := storage.GetObsLogFileName(prefix)
+	if err != nil {
+		log.Error("GetObsLogFileName(%s) failed:%v", jobID, err.Error(), ctx.Data["msgID"])
+		ctx.ServerError("GetObsLogFileName", err)
+		return
+	}
+
+	url, err := storage.GetObsCreateSignedUrlByBucketAndKey(setting.Bucket, key)
+	if err != nil {
+		log.Error("GetObsCreateSignedUrlByBucketAndKey failed: %v", err.Error(), ctx.Data["msgID"])
+		ctx.ServerError("GetObsCreateSignedUrlByBucketAndKey", err)
+		return
+	}
+	ctx.Resp.Header().Set("Cache-Control", "max-age=0")
+	http.Redirect(ctx.Resp, ctx.Req.Request, url, http.StatusMovedPermanently)
+}

routers/routes/routes.go

@@ -1136,6 +1136,7 @@ func RegisterRoutes(m *macaron.Macaron) {
 					m.Post("/stop", cloudbrain.AdminOrOwnerOrJobCreaterRightForTrain, repo.TrainJobStop)
 					m.Post("/del", cloudbrain.AdminOrOwnerOrJobCreaterRightForTrain, repo.TrainJobDel)
 					m.Get("/model_download", cloudbrain.AdminOrJobCreaterRightForTrain, repo.ModelDownload)
+					m.Get("/download_log_file", cloudbrain.AdminOrJobCreaterRightForTrain, repo.TrainJobDownloadLogFile)
 					m.Get("/create_version", reqWechatBind, cloudbrain.AdminOrJobCreaterRightForTrain, repo.TrainJobNewVersion)
 					m.Post("/create_version", reqWechatBind, cloudbrain.AdminOrJobCreaterRightForTrain, bindIgnErr(auth.CreateModelArtsTrainJobForm{}), repo.TrainJobCreateVersion)
 				})

templates/repo/modelarts/trainjob/show.tmpl

@@ -177,6 +177,12 @@
         border: 1px solid #dfe1e6;
     }
 
+    .ti-download-file {
+        display: flex;
+        align-items: center;
+        margin: 0.5rem 0;
+    }
+
     .disabled {
         cursor: default;
         pointer-events: none;
@@ -220,6 +226,7 @@
                 <div class="active section">{{.displayJobName}}</div>
             </div>
         </h4>
+
         {{range $k ,$v := .version_list_task}}
         <div class="ui accordion border-according" id="accordion{{.VersionName}}"
             data-repopath="{{$.RepoRelPath}}/modelarts/train-job" data-jobid="{{.JobID}}"
@@ -479,7 +486,17 @@
                         </div>
                     </div>
                     <div class="ui tab" data-tab="second{{$k}}">
-                        <div style="position: relative;">
+                        <div>
+
+                            <a class='{{if eq .Status "KILLED" "FAILED" "START_FAILED" "STOPPED" "COMPLETED"}}ti-download-file{{else}}disabled{{end}}'
+                                href="{{$.RepoLink}}/modelarts/train-job/{{.JobID}}/download_log_file?version_name={{.VersionName}}">
+                                <i class="ri-download-cloud-2-line"></i>
+                                <span style="margin-left: 0.3rem;">{{$.i18n.Tr "repo.modelarts.download_log"}}</span>
+                            </a>
+
+                        </div>
+                        <div
+                            style="position: relative;border: 1px solid rgba(0,0,0,.2);padding: 0 10px;margin-top: 10px;">
                             <span>
                                 <a title="滚动到顶部" style="position: absolute; right: -32px;cursor: pointer;"
                                     class="log_top" data-version="{{.VersionName}}"><i class="icon-to-top"></i></a>