HPO: DLC mode support nas&oss at same time (#4506)

docs/source/TrainingService/DLCMode.rst

@@ -51,10 +51,10 @@ Use ``examples/trials/mnist-pytorch`` as an example. The NNI config YAML file's
     podCount: 1
     ecsSpec: ecs.c6.large
     region: cn-hangzhou
-    nasDataSourceId: ${your_nas_data_source_id}
     accessKeyId: ${your_ak_id}
     accessKeySecret: ${your_ak_key}
-    nasDataSourceId: ${your_nas_data_source_id}     # NAS datasource ID，e.g., datat56by9n1xt0a
+    nasDataSourceId: ${your_nas_data_source_id}     # NAS datasource ID, e.g., datat56by9n1xt0a
+    ossDataSourceId: ${your_oss_data_source_id}     # OSS datasource ID, in case your data is on oss
     localStorageMountPoint: /home/admin/workspace/  # default NAS path on DSW
     containerStorageMountPoint: /root/data/         # default NAS path on DLC container, change it according your setting
 

docs/source/reference/experiment_config.rst

@@ -584,6 +584,10 @@ Detailed usage can be found `here <../TrainingService/DlcMode.rst>`__.
       - ``str``
       - The NAS datasource id configurated in PAI-DLC side.
 
+    * - ossDataSourceId
+      - ``str``
+      - The OSS datasource id configurated in PAI-DLC side, this is optional.
+
     * - accessKeyId
       - ``str``
       - The accessKeyId of your cloud account.

examples/trials/mnist-pytorch/config_dlc.yml

@@ -17,9 +17,9 @@ trainingService:
   podCount: 1
   ecsSpec: ecs.c6.large
   region: cn-hangzhou
-  nasDataSourceId: ${your_nas_data_source_id}
   accessKeyId: ${your_ak_id}
   accessKeySecret: ${your_ak_key}
   nasDataSourceId: ${your_nas_data_source_id}     # NAS datasource ID，e.g., datat56by9n1xt0a
+  ossDataSourceId: ${your_oss_data_source_id}     # optional, OSS data source id.
   localStorageMountPoint: /home/admin/workspace/  # default NAS path on DSW, MUST provide full path.
   containerStorageMountPoint: /root/data/         # default NAS path on DLC container, change it according your setting

nni/experiment/config/training_services/dlc.py

@@ -2,6 +2,7 @@
 # Licensed under the MIT license.
 
 from dataclasses import dataclass
+from typing import Optional
 
 from ..training_service import TrainingServiceConfig
 
@@ -17,6 +18,7 @@ class DlcConfig(TrainingServiceConfig):
     ecs_spec: str # e.g.,'ecs.c6.large'
     region: str
     nas_data_source_id: str
+    oss_data_source_id: Optional[str] = None
     access_key_id: str
     access_key_secret: str
     local_storage_mount_point: str

ts/nni_manager/common/experimentConfig.ts

@@ -93,6 +93,7 @@ export interface DlcConfig extends TrainingServiceConfig {
     ecsSpec: string;
     region: string;
     nasDataSourceId: string;
+    ossDataSourceId?: string;
     accessKeyId: string;
     accessKeySecret: string;
     localStorageMountPoint: string;

ts/nni_manager/config/dlc/dlcUtil.py

 # Copyright (c) Microsoft Corporation.
 # Licensed under the MIT license.
 
+
 import os
 import sys
-import time
-import json
+import traceback
 from argparse import ArgumentParser
 # ref: https://help.aliyun.com/document_detail/203290.html?spm=a2c4g.11186623.6.727.6f9b5db6bzJh4x
 from alibabacloud_pai_dlc20201203.client import Client
@@ -20,10 +20,12 @@ if __name__ == "__main__":
     parser.add_argument('--ecs_spec', help='ecs spec')
     parser.add_argument('--region', help='region')
     parser.add_argument('--nas_data_source_id', help='nas data_source_id of DLC dataset configuration')
+    parser.add_argument('--oss_data_source_id', help='oss data_source_id of DLC dataset configuration')
     parser.add_argument('--access_key_id', help='access_key_id')
     parser.add_argument('--access_key_secret', help='access_key_secret')
     parser.add_argument('--experiment_name', help='the experiment name')
     parser.add_argument('--user_command', help='user command')
+    parser.add_argument('--log_dir', help='exception log dir')
     args = parser.parse_args()
 
     # init client
@@ -37,10 +39,17 @@ if __name__ == "__main__":
     )
 
     nas_1 = DataSourceItem(
-        data_source_type = 'nas',
+        data_source_type='nas',
         data_source_id=args.nas_data_source_id,
     )
 
+    oss = None
+    if args.oss_data_source_id:
+        oss = DataSourceItem(
+            data_source_type='oss',
+            data_source_id=args.oss_data_source_id,
+        )
+
     # job spec
     spec = JobSpec(
         type=args.type,
@@ -49,15 +58,19 @@ if __name__ == "__main__":
         ecs_spec=args.ecs_spec,
     )
 
+    data_sources = [nas_1]
+    if oss:
+        data_sources = [nas_1, oss]
     req = CreateJobRequest(
         display_name=args.experiment_name,
         job_type=args.job_type,
         job_specs=[spec],
-        data_sources=[nas_1],
+        data_sources=data_sources,
         user_command=args.user_command
     )
 
     # DLC submit
+    try:
         response = client.create_job(req)
         job_id = response.body.job_id
         print('job id: ' + job_id)
@@ -72,3 +85,7 @@ if __name__ == "__main__":
             elif line == 'stop':
                 client.stop_job(job_id)
                 exit(0)
+    except Exception as e:
+        with open(os.path.join(args.log_dir, 'dlc_exception.log'), 'w') as f:
+            f.write('DLC submit Exception: \n')
+            traceback.print_exc(file=f)

ts/nni_manager/training_service/reusable/dlc/dlcClient.ts

@@ -16,11 +16,14 @@ export class DlcClient {
     // e.g., data1e6vg1tu0zi7, to generate it, go to 'Dataset Config' page of DLC
     //       create a NAS data and copy the 'DataSet ConfigurationID'
     public nasDataSourceId: string;
+    public ossDataSourceId: string;
     public accessKeyId: string;
     public accessKeySecret: string;
     public experimentId: string;
     public environmentId: string;
     public userCommand: string;
+    // dlcUtil exception log dir
+    public logDir: string;
     public pythonShellClient: undefined | PythonShell;
 
     constructor(
@@ -36,6 +39,8 @@ export class DlcClient {
         accessKeyId: string,
         accessKeySecret: string,
         userCommand: string,
+        logDir: string,
+        ossDataSourceId?: string,
         ) {
         this.log = getLogger('DlcClient');
         this.type = type;
@@ -46,11 +51,17 @@ export class DlcClient {
         this.image = image;
         this.region = region;
         this.nasDataSourceId = nasDataSourceId;
+        if (ossDataSourceId !== undefined) {
+            this.ossDataSourceId = ossDataSourceId;
+        } else {
+            this.ossDataSourceId = '';
+        }
         this.accessKeyId = accessKeyId;
         this.accessKeySecret = accessKeySecret
         this.experimentId = experimentId;
         this.environmentId = environmentId;
         this.userCommand = userCommand;
+        this.logDir = logDir;
     }
 
     public submit(): Promise<string> {
@@ -67,10 +78,12 @@ export class DlcClient {
                 '--ecs_spec', this.ecsSpec,
                 '--region', this.region,
                 '--nas_data_source_id', this.nasDataSourceId,
+                '--oss_data_source_id', this.ossDataSourceId,
                 '--access_key_id', this.accessKeyId,
                 '--access_key_secret', this.accessKeySecret,
                 '--experiment_name', `nni_exp_${this.experimentId}_env_${this.environmentId}`,
                 '--user_command', this.userCommand,
+                '--log_dir', this.logDir,
               ]
         });
         this.log.debug(this.pythonShellClient.command);

ts/nni_manager/training_service/reusable/environments/dlcEnvironmentService.ts

@@ -116,6 +116,8 @@ export class DlcEnvironmentService extends EnvironmentService {
             this.config.accessKeyId,
             this.config.accessKeySecret,
             environment.command,
+            dlcEnvironment.workingFolder,
+            this.config.ossDataSourceId,
         );
 
         dlcEnvironment.id = await dlcClient.submit();