fix issue

models/cloudbrain.go

@@ -291,6 +291,13 @@ func (task *Cloudbrain) IsRunning() bool {
 		status == string(JobRunning) || status == GrampusStatusRunning
 }
 
+func (task *Cloudbrain) IsUserHasRight(user *User) bool {
+	if user == nil {
+		return false
+	}
+	return user.IsAdmin || user.ID == task.UserID
+}
+
 func ConvertDurationToStr(duration int64) string {
 	if duration <= 0 {
 		return DURATION_STR_ZERO

models/resource_specification.go

@@ -12,6 +12,13 @@ const (
 	SpecOffShelf
 )
 
+type SearchSpecOrderBy int
+
+const (
+	SearchSpecOrderById SearchSpecOrderBy = iota
+	SearchSpecOrder4Standard
+)
+
 type ResourceSpecification struct {
 	ID              int64  `xorm:"pk autoincr"`
 	QueueId         int64  `xorm:"INDEX"`
@@ -85,6 +92,7 @@ type SearchResourceSpecificationOptions struct {
 	Status        int
 	Cluster       string
 	AvailableCode int
+	OrderBy       SearchSpecOrderBy
 }
 
 type SearchResourceBriefSpecificationOptions struct {
@@ -233,10 +241,18 @@ func SearchResourceSpecification(opts SearchResourceSpecificationOptions) (int64
 		return 0, nil, err
 	}
 
+	var orderby = ""
+	switch opts.OrderBy {
+	case SearchSpecOrder4Standard:
+		orderby = "resource_queue.compute_resource asc,resource_queue.acc_card_type asc,resource_specification.acc_cards_num asc,resource_specification.cpu_cores asc,resource_specification.mem_gi_b asc,resource_specification.share_mem_gi_b asc"
+	default:
+		orderby = "resource_specification.id desc"
+	}
+
 	r := make([]ResourceSpecAndQueue, 0)
 	err = x.Where(cond).
 		Join("INNER", "resource_queue", "resource_queue.ID = resource_specification.queue_id").
-		Desc("resource_specification.id").
+		OrderBy(orderby).
 		Limit(opts.PageSize, (opts.Page-1)*opts.PageSize).
 		Unscoped().Find(&r)
 	if err != nil {

modules/grampus/resty.go

@@ -245,6 +245,32 @@ func GetTrainJobLog(jobID string) (string, error) {
 	return logContent, nil
 }
 
+func GetGrampusMetrics(jobID string) (models.GetTrainJobMetricStatisticResult, error) {
+	checkSetting()
+	client := getRestyClient()
+	var result models.GetTrainJobMetricStatisticResult
+	res, err := client.R().
+		SetAuthToken(TOKEN).
+		Get(HOST + urlTrainJob + "/" + jobID + "/task/0/replica/0/metrics")
+
+	if err != nil {
+		return result, fmt.Errorf("resty GetTrainJobLog: %v", err)
+	}
+	if err = json.Unmarshal([]byte(res.String()), &result); err != nil {
+		log.Error("GetGrampusMetrics json.Unmarshal failed(%s): %v", res.String(), err.Error())
+		return result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
+	}
+	if res.StatusCode() != http.StatusOK {
+		log.Error("Call GrampusMetrics failed(%d):%s(%s)", res.StatusCode(), result.ErrorCode, result.ErrorMsg)
+		return result, fmt.Errorf("Call GrampusMetrics failed(%d):%d(%s)", res.StatusCode(), result.ErrorCode, result.ErrorMsg)
+	}
+	if !result.IsSuccess {
+		log.Error("GetGrampusMetrics(%s) failed", jobID)
+		return result, fmt.Errorf("GetGrampusMetrics failed:%s", result.ErrorMsg)
+	}
+	return result, nil
+}
+
 func StopJob(jobID string) (*models.GrampusStopJobResponse, error) {
 	checkSetting()
 	client := getRestyClient()

routers/admin/resources.go

@@ -127,6 +127,7 @@ func GetResourceSpecificationList(ctx *context.Context) {
 		Status:        status,
 		Cluster:       cluster,
 		AvailableCode: available,
+		OrderBy:       models.SearchSpecOrderById,
 	})
 	if err != nil {
 		log.Error("GetResourceSpecificationList error.%v", err)

routers/api/v1/api.go

@@ -1048,6 +1048,7 @@ func RegisterRoutes(m *macaron.Macaron) {
 							m.Get("", repo.GetModelArtsTrainJobVersion)
 							m.Post("/stop_version", cloudbrain.AdminOrOwnerOrJobCreaterRightForTrain, repo_ext.GrampusStopJob)
 							m.Get("/log", repo_ext.GrampusGetLog)
+							m.Get("/metrics", repo_ext.GrampusMetrics)
 							m.Get("/download_log", cloudbrain.AdminOrJobCreaterRightForTrain, repo_ext.GrampusDownloadLog)
 						})
 					})

routers/api/v1/repo/cloudbrain.go

@@ -661,14 +661,20 @@ func CloudbrainGetLog(ctx *context.APIContext) {
 	if ctx.Data["existStr"] != nil && result["Lines"].(int) < 50 {
 		content = content + ctx.Data["existStr"].(string)
 	}
+	logFileName := result["FileName"]
+
+	//Logs can only be downloaded if the file exists
+	//and the current user is an administrator or the creator of the task
+	canLogDownload := logFileName != nil && logFileName != "" && job.IsUserHasRight(ctx.User)
+
 	re := map[string]interface{}{
 		"JobID":          ID,
-		"LogFileName":    result["FileName"],
+		"LogFileName":    logFileName,
 		"StartLine":      result["StartLine"],
 		"EndLine":        result["EndLine"],
 		"Content":        content,
 		"Lines":          result["Lines"],
-		"CanLogDownload": result["FileName"] != "",
+		"CanLogDownload": canLogDownload,
 		"StartTime":      job.StartTime,
 	}
 	//result := CloudbrainGetLogByJobId(job.JobID, job.JobName)

routers/api/v1/repo/modelarts.go

@@ -281,15 +281,6 @@ func TrainJobGetLog(ctx *context.APIContext) {
 		return
 	}
 
-	prefix := strings.TrimPrefix(path.Join(setting.TrainJobModelPath, task.JobName, modelarts.LogPath, versionName), "/") + "/job"
-	_, err = storage.GetObsLogFileName(prefix)
-	var canLogDownload bool
-	if err != nil {
-		canLogDownload = false
-	} else {
-		canLogDownload = true
-	}
-
 	ctx.Data["log_file_name"] = resultLogFile.LogFileList[0]
 
 	ctx.JSON(http.StatusOK, map[string]interface{}{
@@ -299,11 +290,23 @@ func TrainJobGetLog(ctx *context.APIContext) {
 		"EndLine":        result.EndLine,
 		"Content":        result.Content,
 		"Lines":          result.Lines,
-		"CanLogDownload": canLogDownload,
+		"CanLogDownload": canLogDownload(ctx.User, task),
 		"StartTime":      task.StartTime,
 	})
 }
 
+func canLogDownload(user *models.User, task *models.Cloudbrain) bool {
+	if task == nil || !task.IsUserHasRight(user) {
+		return false
+	}
+	prefix := strings.TrimPrefix(path.Join(setting.TrainJobModelPath, task.JobName, modelarts.LogPath, task.VersionName), "/") + "/job"
+	_, err := storage.GetObsLogFileName(prefix)
+	if err != nil {
+		return false
+	}
+	return true
+}
+
 func trainJobGetLogContent(jobID string, versionID int64, baseLine string, order string, lines int) (*models.GetTrainJobLogFileNamesResult, *models.GetTrainJobLogResult, error) {
 
 	resultLogFile, err := modelarts.GetTrainJobLogFileNames(jobID, strconv.FormatInt(versionID, 10))

routers/repo/grampus.go

@@ -940,15 +940,14 @@ func GrampusGetLog(ctx *context.Context) {
 	content, err := grampus.GetTrainJobLog(job.JobID)
 	if err != nil {
 		log.Error("GetTrainJobLog failed: %v", err, ctx.Data["MsgID"])
-		ctx.ServerError(err.Error(), err)
+		ctx.JSON(http.StatusOK, map[string]interface{}{
+			"JobName":        job.JobName,
+			"Content":        "",
+			"CanLogDownload": false,
+		})
 		return
 	}
-	var canLogDownload bool
-	if err != nil {
-		canLogDownload = false
-	} else {
-		canLogDownload = true
-	}
+	canLogDownload := err == nil && job.IsUserHasRight(ctx.User)
 	ctx.JSON(http.StatusOK, map[string]interface{}{
 		"JobName":        job.JobName,
 		"Content":        content,
@@ -958,6 +957,28 @@ func GrampusGetLog(ctx *context.Context) {
 	return
 }
 
+func GrampusMetrics(ctx *context.Context) {
+	jobID := ctx.Params(":jobid")
+	job, err := models.GetCloudbrainByJobID(jobID)
+	if err != nil {
+		log.Error("GetCloudbrainByJobID failed: %v", err, ctx.Data["MsgID"])
+		ctx.ServerError(err.Error(), err)
+		return
+	}
+
+	result, err := grampus.GetGrampusMetrics(job.JobID)
+	if err != nil {
+		log.Error("GetTrainJobLog failed: %v", err, ctx.Data["MsgID"])
+	}
+	ctx.JSON(http.StatusOK, map[string]interface{}{
+		"JobID":       jobID,
+		"Interval":    result.Interval,
+		"MetricsInfo": result.MetricsInfo,
+	})
+
+	return
+}
+
 func generateCommand(repoName, processorType, codeRemotePath, dataRemotePath, bootFile, paramSrc, outputRemotePath, datasetName, pretrainModelPath, pretrainModelFileName, modelRemoteObsUrl string) (string, error) {
 	var command string
 

services/cloudbrain/cloudbrainTask/sync_status.go

@@ -14,7 +14,7 @@ import (
 var noteBookOKMap = make(map[int64]int, 20)
 
 //if a task notebook url can get  two times,  the notebook can browser.
-const successfulCount = 2
+const successfulCount = 3
 
 func SyncCloudBrainOneStatus(task *models.Cloudbrain) (*models.Cloudbrain, error) {
 	jobResult, err := cloudbrain.GetJob(task.JobID)

services/cloudbrain/resource/resource_specification.go

@@ -138,6 +138,7 @@ func GetResourceSpecificationList(opts models.SearchResourceSpecificationOptions
 func GetAllDistinctResourceSpecification(opts models.SearchResourceSpecificationOptions) (*models.ResourceSpecAndQueueListRes, error) {
 	opts.Page = 0
 	opts.PageSize = 1000
+	opts.OrderBy = models.SearchSpecOrder4Standard
 	_, r, err := models.SearchResourceSpecification(opts)
 	if err != nil {
 		return nil, err

templates/repo/cloudbrain/trainjob/show.tmpl

@@ -284,7 +284,8 @@
                 <div class="content-pad">
                     <div class="ui pointing secondary menu" style="border-bottom: 1px solid rgba(34,36,38,.15);">
                         <a class="active item"
-                            data-tab="first{{$k}}">{{$.i18n.Tr "repo.modelarts.train_job.config"}}</a>
+                            data-tab="first{{$k}}">{{$.i18n.Tr "repo.modelarts.train_job.config"}}</a>       
+
                         <a class="item log_bottom" data-tab="third{{$k}}"
                             data-version="{{.VersionName}}">{{$.i18n.Tr "repo.modelarts.log"}}</a>                       
                         <a class="item load-model-file" data-tab="four{{$k}}" data-gpu-flag="true" data-download-flag="{{$.canDownload}}" data-path="{{$.RepoLink}}/cloudbrain/train-job/{{.JobID}}/model_list" data-version="{{.VersionName}}" data-parents="" data-filename="" data-init="init" >{{$.i18n.Tr "repo.model_download"}}</a>

templates/repo/grampus/trainjob/show.tmpl

@@ -238,11 +238,8 @@
                             <span>
                                 <div style="float: right;">
                                     {{$.CsrfTokenHtml}}
-                                    
                                 </div>
-                               
                                 <div class="ac-display-inblock title_text acc-margin-bottom">
-
                                     <span class="cti-mgRight-sm">{{TimeSinceUnix1 .CreatedUnix}}</span>
                                     <span class="cti-mgRight-sm">
                                         {{$.i18n.Tr "repo.modelarts.current_version"}}：{{.VersionName}}</span>
@@ -260,7 +257,6 @@
                                     <span class="refresh-status" data-tooltip="刷新" style="cursor: pointer;" data-inverted="" data-version="{{.VersionName}}">
                                         <i class="redo icon redo-color"></i>
                                     </span>
-
                                 </div>
                                  <div style="float: right;">
                                     {{if and ($.canDownload) (ne .Status "WAITING") ($.Permission.CanWrite $.UnitTypeModelManage) }}
@@ -269,7 +265,6 @@
                                     {{else}}
                                     <a class="ti-action-menu-item disabled" id="{{.VersionName}}-create-model">{{$.i18n.Tr "repo.modelarts.create_model"}}</a>
                                     {{end}}
-
                                 </div>
                             </span>
                         </span>
@@ -282,6 +277,9 @@
 
                         <a class="active item" data-tab="first{{$k}}">{{$.i18n.Tr "repo.modelarts.train_job.config"}}</a>
                         <a class="item log_bottom" data-tab="second{{$k}}" data-version="{{.VersionName}}">{{$.i18n.Tr "repo.modelarts.log"}}</a>
+                        {{ if eq $.Spec.ComputeResource "NPU"}}
+                        <a class="item metric_chart" data-tab="four{{$k}}" data-version="{{.VersionName}}" data-path="{{$.RepoRelPath}}/grampus/train-job/{{.JobID}}/metrics">{{$.i18n.Tr "cloudbrain.resource_use"}}</a>
+                        {{end}}
                         <a class="item load-model-file" data-tab="third{{$k}}" data-download-flag="{{$.canDownload}}" data-path="{{$.RepoLink}}/modelarts/train-job/{{.JobID}}/model_list" data-version="{{.VersionName}}" data-parents="" data-filename="" data-init="init" >{{$.i18n.Tr "repo.model_download"}}</a>
                     </div>
                     <div class="ui tab active" data-tab="first{{$k}}">
@@ -564,6 +562,14 @@
 
                         </div>
 
+                    </div>
+                    <div class="ui tab" data-tab="four{{$k}}" style="position: relative;">
+                        <i class="ri-refresh-line metric_chart"
+                            style="position: absolute;right: 25%;color:#3291f8;z-index:99;cursor: pointer;"
+                            data-version="{{.VersionName}}"></i>
+                        <div id="metric-{{.VersionName}}" style="height: 260px;width: 870px;">
+                        </div>
+                        
                     </div>
                     <div class="ui tab" data-tab="third{{$k}}">
                         <input type="hidden" name="model{{.VersionName}}" value="-1">

templates/repo/modelarts/trainjob/show.tmpl

@@ -321,7 +321,7 @@
                             data-tab="first{{$k}}">{{$.i18n.Tr "repo.modelarts.train_job.config"}}</a>
                         <a class="item log_bottom" data-tab="second{{$k}}"
                             data-version="{{.VersionName}}">{{$.i18n.Tr "repo.modelarts.log"}}</a>
-                        <a class="item metric_chart" data-tab="four{{$k}}" data-version="{{.VersionName}}">{{$.i18n.Tr "cloudbrain.resource_use"}}</a>
+                        <a class="item metric_chart" data-tab="four{{$k}}" data-version="{{.VersionName}}" data-path="{{$.RepoRelPath}}/modelarts/train-job/{{.JobID}}/metric_statistics?version_name={{.VersionName}}&statistic_type=each&metrics=">{{$.i18n.Tr "cloudbrain.resource_use"}}</a>
                         <a class="item load-model-file" data-tab="third{{$k}}" data-download-flag="{{$.canDownload}}" data-path="{{$.RepoLink}}/modelarts/train-job/{{.JobID}}/model_list" data-version="{{.VersionName}}" data-parents="" data-filename="" data-init="init" >{{$.i18n.Tr "repo.model_download"}}</a>
                     </div>
                     <div class="ui tab active" data-tab="first{{$k}}">

web_src/js/index.js

@@ -5071,12 +5071,7 @@ function initcreateRepo() {
 initcreateRepo();
 
 function initChartsNpu() {
-  const url = window.location.href;
-  const urlArr = url.split("/");
-  let userName = urlArr.slice(-5)[0];
-  let repoPath = urlArr.slice(-4)[0];
-  let jobID = urlArr.slice(-1)[0];
-
+  const repoPath = $('.metric_chart').data('path')
   let options = {
     legend: {
       data: [],
@@ -5127,7 +5122,7 @@ function initChartsNpu() {
       document.getElementById(`metric-${versionName}`)
     );
     $.get(
-      `${window.config.AppSubUrl}/api/v1/repos/${userName}/${repoPath}/modelarts/train-job/${jobID}/metric_statistics?version_name=${versionName}&statistic_type=each&metrics=`,
+      `${window.config.AppSubUrl}/api/v1/repos/${repoPath}`,
       (res) => {
         let filterDta = res.MetricsInfo.filter((item) => {
           return ![