Fix OpenPAI IT (#4057)

microsoft · Aug 11, 2021 · 76152d4 · 76152d4
1 parent 56da3c3
commit 76152d4
Show file tree

Hide file tree

Showing 5 changed files with 15 additions and 10 deletions.
diff --git a/pipelines/integration-test-openpai-linux.yml b/pipelines/integration-test-openpai-linux.yml
@@ -64,10 +64,11 @@ jobs:
           --nni_docker_image nnidev/nni-nightly \
           --pai_storage_config_name confignfs-data \
           --pai_token $(pai_token) \
-          --nni_manager_nfs_mount_path /home/quzha/mnt-pai-ne/shinyang3 \
-          --container_nfs_mount_path /mnt/confignfs-data/shinyang3 \
+          --nni_manager_nfs_mount_path $(nni_manager_nfs_mount_path) \
+          --container_nfs_mount_path $(container_nfs_mount_path) \
           --nni_manager_ip $(manager_ip) \
-          --vc nni
+          --vc nni \
+          --debug true
       python3 nni_test/nnitest/run_tests.py --config config/integration_tests.yml --ts pai
     displayName: Integration test
 
@@ -82,8 +83,8 @@ jobs:
           --nni_docker_image nnidev/nni-nightly \
           --pai_storage_config_name confignfs-data \
           --pai_token $(pai_token) \
-          --nni_manager_nfs_mount_path /home/quzha/mnt-pai-ne/shinyang3 \
-          --container_nfs_mount_path /mnt/confignfs-data/shinyang3 \
+          --nni_manager_nfs_mount_path $(nni_manager_nfs_mount_path) \
+          --container_nfs_mount_path $(container_nfs_mount_path) \
           --nni_manager_ip $(manager_ip) \
           --vc nni
       python3 nni_test/nnitest/run_tests.py --config config/integration_tests.yml --ts pai

diff --git a/test/nni_test/nnitest/generate_ts_config.py b/test/nni_test/nnitest/generate_ts_config.py
@@ -33,6 +33,8 @@ def update_training_service_config(args):
             config[args.ts]['trial']['paiStorageConfigName'] = args.pai_storage_config_name
         if args.vc is not None:
             config[args.ts]['trial']['virtualCluster'] = args.vc
+        if args.debug is not None:
+            config[args.ts]['debug'] = args.debug.lower() == 'true'
     elif args.ts == 'kubeflow':
         if args.nfs_server is not None:
             config[args.ts]['kubeflowConfig']['nfs']['server'] = args.nfs_server
@@ -146,6 +148,7 @@ def update_training_service_config(args):
     parser.add_argument("--pai_storage_config_name", type=str)
     parser.add_argument("--nni_manager_nfs_mount_path", type=str)
     parser.add_argument("--container_nfs_mount_path", type=str)
+    parser.add_argument("--debug", type=str)
     # args for kubeflow and frameworkController
     parser.add_argument("--nfs_path", type=str)
     parser.add_argument("--keyvault_vaultname", type=str)

diff --git a/ts/nni_manager/common/experimentConfig.ts b/ts/nni_manager/common/experimentConfig.ts
@@ -228,4 +228,4 @@ export function flattenConfig<T>(config: ExperimentConfig, platform: string): T
         Object.assign(flattened, config.trainingService);
     }
     return <T>flattened;
-}
+}
diff --git a/ts/nni_manager/training_service/pai/paiTrainingService.ts b/ts/nni_manager/training_service/pai/paiTrainingService.ts
@@ -70,6 +70,7 @@ class PAITrainingService implements TrainingService {
         this.paiTokenUpdateInterval = 7200000; //2hours
         this.log.info('Construct paiBase training service.');
         this.config = flattenConfig(config, 'openpai');
+        this.versionCheck = !this.config.debug;
         this.paiJobRestServer = new PAIJobRestServer(this);
         this.paiToken = this.config.token;
         this.protocol = this.config.host.toLowerCase().startsWith('https://') ? 'https' : 'http';
@@ -78,7 +79,7 @@ class PAITrainingService implements TrainingService {
 
     private async copyTrialCode(): Promise<void> {
         await validateCodeDir(this.config.trialCodeDirectory);
-        const nniManagerNFSExpCodeDir = path.join(this.config.trialCodeDirectory, this.experimentId, 'nni-code');
+        const nniManagerNFSExpCodeDir = path.join(this.config.localStorageMountPoint, this.experimentId, 'nni-code');
         await execMkdir(nniManagerNFSExpCodeDir);
         this.log.info(`Starting copy codeDir data from ${this.config.trialCodeDirectory} to ${nniManagerNFSExpCodeDir}`);
         await execCopydir(this.config.trialCodeDirectory, nniManagerNFSExpCodeDir);

diff --git a/ts/nni_manager/training_service/reusable/routerTrainingService.ts b/ts/nni_manager/training_service/reusable/routerTrainingService.ts
@@ -26,11 +26,11 @@ class RouterTrainingService implements TrainingService {
         const instance = new RouterTrainingService();
         instance.log = getLogger('RouterTrainingService');
         const platform = Array.isArray(config.trainingService) ? 'hybrid' : config.trainingService.platform;
-        if (platform === 'remote' && !(<RemoteConfig>config.trainingService).reuseMode) {
+        if (platform === 'remote' && (<RemoteConfig>config.trainingService).reuseMode === false) {
             instance.internalTrainingService = new RemoteMachineTrainingService(config);
-        } else if (platform === 'openpai' && !(<OpenpaiConfig>config.trainingService).reuseMode) {
+        } else if (platform === 'openpai' && (<OpenpaiConfig>config.trainingService).reuseMode === false) {
             instance.internalTrainingService = new PAITrainingService(config);
-        } else if (platform === 'kubeflow' && !(<KubeflowConfig>config.trainingService).reuseMode) {
+        } else if (platform === 'kubeflow' && (<KubeflowConfig>config.trainingService).reuseMode === false) {
             instance.internalTrainingService = new KubeflowTrainingService();
         } else {
             instance.internalTrainingService = await TrialDispatcher.construct(config);
-Original file line number
+Diff line change
@@ Expand Up @@
             Object.assign(flattened, config.trainingService);
         }
         return <T>flattened;
-    }
+    }