提交代码

modules/grampus/grampus.go

@@ -29,7 +29,7 @@ const (
 	BucketRemote       = "grampus"
 	RemoteModelPath    = "/output/" + models.ModelSuffix
 	autoStopDurationMs = 4 * 60 * 60 * 1000
-	CommandGpuDebug    = "mkdir -p /dataset;%s! [ -x \"$(command -v jupyter)\" ] && pip install jupyterlab==3 -i https://pypi.tuna.tsinghua.edu.cn/simple;jupyter lab --no-browser --ip=0.0.0.0 --allow-root --notebook-dir='/code' --port=$OCTOPUS_NOTEBOOK_PORT --LabApp.token='' --LabApp.allow_origin='*' --LabApp.base_url=$OCTOPUS_NOTEBOOK_BASE_URL;"
+	CommandGpuDebug    = "mkdir -p /dataset;%s! [ -x \"$(command -v jupyter)\" ] && pip install jupyterlab==3 -i https://pypi.tuna.tsinghua.edu.cn/simple;jupyter lab --ServerApp.shutdown_no_activity_timeout=%s --TerminalManager.cull_inactive_timeout=%s --TerminalManager.cull_interval=%s --MappingKernelManager.cull_idle_timeout=%s --MappingKernelManager.cull_interval=%s --MappingKernelManager.cull_connected=True --MappingKernelManager.cull_busy=True --no-browser --ip=0.0.0.0 --allow-root --notebook-dir='/code' --port=$OCTOPUS_NOTEBOOK_PORT --LabApp.token='' --LabApp.allow_origin='*' --LabApp.base_url=$OCTOPUS_NOTEBOOK_BASE_URL;"
 )
 
 var (
@@ -206,7 +206,7 @@ func GenerateNotebookJob(ctx *context.Context, req *GenerateNotebookJobReq) (job
 			ReadOnly:      false,
 			ContainerPath: cloudbrain.CodeMountPath,
 		}
-		req.Command = fmt.Sprintf(CommandGpuDebug, cpCommand)
+		req.Command = fmt.Sprintf(CommandGpuDebug, cpCommand, setting.CullIdleTimeout, setting.CullIdleTimeout, setting.CullInterval, setting.CullIdleTimeout, setting.CullInterval)
 		log.Info("debug command:" + req.Command)
 
 	}

options/locale/locale_en-US.ini

@@ -1341,6 +1341,9 @@ modelconvert.manage.create_error2=Only one running model transformation task can
 modelconvert.manage.model_not_exist=The model in the task does not exist or has been deleted.
 modelconvert.manage.no_operate_right=You have no right to do the operation.
 
+debug.manage.model_not_exist=The model in the task does not exist or has been deleted, please create a new debug job.
+debug.manage.dataset_not_exist=The part of datasets in the task does not exist or has been deleted, please create a new debug job.
+
 grampus.train_job.ai_center = AI Center
 grampus.dataset_path_rule = The code is storaged in /cache/code;the dataset is storaged in /cache/dataset;and please put your model into /cache/output, then you can download it online。
 grampus.gpu_dataset_path_rule = The code is storaged in /tmp/code;the dataset is storaged in /tmp/dataset;and please put your model into /tmp/output, then you can download it online。

options/locale/locale_zh-CN.ini

@@ -1355,6 +1355,10 @@ modelconvert.manage.create_error2=只能创建一个正在运行的模型转换
 modelconvert.manage.model_not_exist=任务中选择的模型不存在或者已被删除。
 modelconvert.manage.no_operate_right=您没有操作权限。
 
+
+debug.manage.model_not_exist=任务中选择的模型不存在或者已被删除，请新建调试任务。
+debug.manage.dataset_not_exist=任务中选择的部分数据集不存在或者已被删除，请新建调试任务。
+
 grampus.train_job.ai_center=智算中心
 grampus.dataset_path_rule = 训练脚本存储在/cache/code中，数据集存储在/cache/dataset中，训练输出请存储在/cache/output中以供后续下载。
 grampus.gpu_dataset_path_rule = 训练脚本存储在/tmp/code中，数据集存储在/tmp/dataset中，训练输出请存储在/tmp/output中以供后续下载。

routers/repo/cloudbrain.go

@@ -706,7 +706,13 @@ func CloudBrainRestart(ctx *context.Context) {
 		}
 		if !HasModelFile(task) {
 			resultCode = "-1"
-			errorMsg = ctx.Tr("repo.modelconvert.manage.model_not_exist")
+			errorMsg = ctx.Tr("repo.debug.manage.model_not_exist")
+			break
+		}
+
+		if hasDatasetDeleted(task) {
+			resultCode = "-1"
+			errorMsg = ctx.Tr("repo.debug.manage.dataset_not_exist")
 			break
 		}
 
@@ -729,6 +735,14 @@ func CloudBrainRestart(ctx *context.Context) {
 	})
 }
 
+func hasDatasetDeleted(task *models.Cloudbrain) bool {
+	if task.Uuid == "" {
+		return false
+	}
+	uuids := strings.Split(task.Uuid, ";")
+	attachs, _ := models.GetAttachmentsByUUIDs(uuids)
+	return len(attachs) < len(uuids)
+}
 
 func HasModelFile(task *models.Cloudbrain) bool {
 	if task.PreTrainModelUrl == "" {

routers/repo/grampus.go

@@ -1670,7 +1670,11 @@ func GrampusNotebookRestart(ctx *context.Context) {
 		}
 
 		if !HasModelFile(task) { //使用预训练模型训练
-			errorMsg = ctx.Tr("repo.modelconvert.manage.model_not_exist")
+			errorMsg = ctx.Tr("repo.debug.manage.model_not_exist")
+			break
+		}
+		if hasDatasetDeleted(task) {
+			errorMsg = ctx.Tr("repo.debug.manage.dataset_not_exist")
 			break
 		}
 

routers/repo/modelarts.go

@@ -525,7 +525,12 @@ func NotebookRestart(ctx *context.Context) {
 			break
 		}
 		if !HasModelFile(task) { //使用预训练模型训练
-			errorMsg = ctx.Tr("repo.modelconvert.manage.model_not_exist")
+			errorMsg = ctx.Tr("repo.debug.manage.model_not_exist")
+			break
+		}
+
+		if hasDatasetDeleted(task) {
+			errorMsg = ctx.Tr("repo.debug.manage.dataset_not_exist")
 			break
 		}