Add shared storage integration test (#3455)

nni/experiment/config/common.py

@@ -46,6 +46,7 @@ class CustomAlgorithmConfig(_AlgorithmConfig):
 class TrainingServiceConfig(ConfigBase):
     platform: str
 
+@dataclass(init=False)
 class SharedStorageConfig(ConfigBase):
     storage_type: str
     local_mount_point: str

pipelines/integration-test-remote-l2l.yml

@@ -87,27 +87,30 @@ jobs:
       cd test
       python3 nni_test/nnitest/generate_ts_config.py \
           --ts remote \
-          --remote_reuse false \
+          --remote_reuse true \
           --remote_user nni \
           --remote_host $(worker_ip) \
           --remote_port $(docker_port) \
           --remote_pwd $(password_in_docker) \
-          --nni_manager_ip $(manager_ip)
+          --nni_manager_ip $(manager_ip) \
+          --azurestoragetoken $(azureblob_token_test) \
+          --nfs_server $(NFS_IP)
       python3 nni_test/nnitest/run_tests.py --config config/integration_tests.yml --ts remote
-    displayName: Integration test
+    displayName: Integration test (reuse mode)
 
   - script: |
       cd test
       python3 nni_test/nnitest/generate_ts_config.py \
           --ts remote \
-          --remote_reuse true \
+          --remote_reuse false \
           --remote_user nni \
           --remote_host $(worker_ip) \
           --remote_port $(docker_port) \
           --remote_pwd $(password_in_docker) \
           --nni_manager_ip $(manager_ip)
       python3 nni_test/nnitest/run_tests.py --config config/integration_tests.yml --ts remote
-    displayName: Integration test (reuse mode)
+    displayName: Integration test
+
 
   - task: SSH@0
     inputs:

test/config/integration_tests.yml

@@ -34,6 +34,34 @@ testCases:
   # check status of experiment before calling validator
   experimentStatusCheck: True
 
+- name: shared-storage-remote-azureblob
+  configFile: test/config/sharedstorage_test/config_sharedstorage_remote_azureblob.yml
+  config:
+    sharedStorage:
+      localMountPoint: /tmp/nnimount/testlocalrootpath
+      remoteMountPoint: /tmp/nnimount/testremoterootpath
+      storageAccountName: nennistorage
+      storageAccountKey: $(azureblob_token_test)
+      containerName: sharedstorage
+  validator:
+    class: FileExistValidator
+    kwargs:
+      rootpath: /tmp/nnimount/testlocalrootpath
+
+# TODO: Enable this case after nfs server is ready
+#- name: shared-storage-remote-nfs
+#  configFile: test/config/sharedstorage_test/config_sharedstorage_remote_nfs.yml
+#  config:
+#    sharedStorage:
+#      localMountPoint: /tmp/nnimount/testlocalrootpath
+#      remoteMountPoint: /tmp/nnimount/testremoterootpath
+#      nfsServer: $(NFS_IP)
+#      exportedDirectory: /home/nni/mnt/
+#  validator:
+#    class: FileExistValidator
+#    kwargs:
+#      rootpath: /tmp/nnimount/testlocalrootpath
+
 - name: sklearn-regression
   configFile: test/config/examples/sklearn-regression.yml
 
@@ -227,4 +255,3 @@ testCases:
 #########################################################################
 - name: customized-tuners-demotuner
   configFile: test/config/customized_tuners/demotuner-sklearn-classification.yml
-

test/config/sharedstorage_test/config_sharedstorage_remote_azureblob.yml

+authorName: default
+experimentName: example_mnist
+trialConcurrency: 1
+maxExecDuration: 1h
+maxTrialNum: 1
+trainingServicePlatform: remote
+searchSpacePath: config_sharedstorage_search_space.json
+#choice: true, false
+useAnnotation: false
+nniManagerIp: 127.0.0.1
+tuner:
+  #choice: TPE, Random, Anneal, Evolution, BatchTuner, MetisTuner, GPTuner
+  #SMAC (SMAC should be installed through nnictl)
+  builtinTunerName: TPE
+  classArgs:
+    #choice: maximize, minimize
+    optimize_mode: maximize
+trial:
+  command: python3 config_sharedstorage_trial.py
+  codeDir: .
+  gpuNum: 0
+sharedStorage:
+  storageType: AzureBlob
+  localMountPoint: ${your/local/mount/point}
+  remoteMountPoint: ${your/remote/mount/point}
+  storageAccountName: ${replace_to_your_storageAccountName}
+  storageAccountKey: ${replace_to_your_storageAccountKey}
+  # If you did not set storageAccountKey, you need use `az login` with Azure CLI at first and set resourceGroupName.
+  # resourceGroupName: ${replace_to_your_resourceGroupName}
+  containerName: ${replace_to_your_containerName}
+  # usermount means you have already mount this storage on localMountPoint
+  # nnimount means nni will try to mount this storage on localMountPoint
+  # nomount means storage will not mount in local machine, will support partial storages in the future 
+  localMounted: nnimount
+#machineList can be empty if the platform is local
+machineList:
+  - ip: 10.1.1.1
+    username: bob
+    passwd: bob123
+    #port can be skip if using default ssh port 22
+    #port: 22
+remoteConfig:
+  reuse: true
\ No newline at end of file

test/config/sharedstorage_test/config_sharedstorage_remote_nfs.yml

+authorName: default
+experimentName: example_mnist
+trialConcurrency: 1
+maxExecDuration: 1h
+maxTrialNum: 1
+trainingServicePlatform: remote
+searchSpacePath: config_sharedstorage_search_space.json
+#choice: true, false
+useAnnotation: false
+nniManagerIp: 127.0.0.1
+tuner:
+  #choice: TPE, Random, Anneal, Evolution, BatchTuner, MetisTuner, GPTuner
+  #SMAC (SMAC should be installed through nnictl)
+  builtinTunerName: TPE
+  classArgs:
+    #choice: maximize, minimize
+    optimize_mode: maximize
+trial:
+  command: python3 config_sharedstorage_trial.py
+  codeDir: .
+  gpuNum: 0
+sharedStorage:
+  storageType: NFS
+  localMountPoint: ${your/local/mount/point}
+  remoteMountPoint: ${your/remote/mount/point}
+  nfsServer: ${nfs-server-ip}
+  exportedDirectory: ${nfs/exported/directory}
+  # usermount means you have already mount this storage on localMountPoint
+  # nnimount means nni will try to mount this storage on localMountPoint
+  # nomount means storage will not mount in local machine, will support partial storages in the future 
+  localMounted: nnimount
+#machineList can be empty if the platform is local
+machineList:
+  - ip: 10.1.1.1
+    username: bob
+    passwd: bob123
+    #port can be skip if using default ssh port 22
+    #port: 22
+remoteConfig:
+  reuse: true
\ No newline at end of file

test/config/sharedstorage_test/config_sharedstorage_search_space.json

+{
+    "dropout_rate":{"_type":"uniform","_value":[0.5, 0.9]},
+    "conv_size":{"_type":"choice","_value":[2,3,5,7]},
+    "hidden_size":{"_type":"choice","_value":[124, 512, 1024]},
+    "batch_size": {"_type":"choice", "_value": [16, 32]},
+    "learning_rate":{"_type":"choice","_value":[0.0001, 0.001, 0.01, 0.1]}
+}

test/config/sharedstorage_test/config_sharedstorage_trial.py

+"""
+A deep MNIST classifier using convolutional layers.
+
+This file is a modification of the official pytorch mnist example:
+https://github.com/pytorch/examples/blob/master/mnist/main.py
+"""
+import os
+import logging
+import nni
+logger = logging.getLogger('mnist_AutoML')
+if __name__ == '__main__':
+    try:
+        logger.debug(os.environ.get('NNI_OUTPUT_DIR'))
+        filename = os.path.join(os.environ.get('NNI_OUTPUT_DIR'), 'checkingfile.txt')
+        f = open(filename, "a")
+        
+        tuner_params = nni.get_next_parameter()
+        f.write(str(tuner_params))
+        nni.report_final_result(1)
+        
+        f.close()
+    except Exception as exception:
+        logger.exception(exception)
+        raise

test/config/training_service.yml

@@ -87,6 +87,9 @@ remote:
     port:
     username:
   trainingServicePlatform: remote
+  sharedStorage:
+    storageAccountKey:
+    nfsServer:
 hybrid:
   maxExecDuration: 15m
   nniManagerIp:

test/nni_test/nnitest/generate_ts_config.py

@@ -74,6 +74,10 @@ def update_training_service_config(args):
             config[args.ts]['machineList'][0]['passwd'] = args.remote_pwd
         if args.remote_reuse is not None:
             config[args.ts]['remoteConfig']['reuse'] = args.remote_reuse.lower() == 'true'
+        if args.azurestoragetoken is not None:
+            config[args.ts]['sharedStorage']['storageAccountKey'] = args.azurestoragetoken
+        if args.nfs_server is not None:
+            config[args.ts]['sharedStorage']['nfsServer'] = args.nfs_server
     elif args.ts == 'adl':
         if args.nni_docker_image is not None:
             config[args.ts]['trial']['image'] = args.nni_docker_image
@@ -118,6 +122,8 @@ if __name__ == '__main__':
     parser.add_argument("--config_version", type=str, choices=['v1', 'v2'], default='v1')
     parser.add_argument("--nni_docker_image", type=str)
     parser.add_argument("--nni_manager_ip", type=str)
+    parser.add_argument("--azurestoragetoken", type=str)
+    parser.add_argument("--nfs_server", type=str)
     # args for PAI
     parser.add_argument("--pai_user", type=str)
     parser.add_argument("--pai_pwd", type=str)
@@ -131,7 +137,6 @@ if __name__ == '__main__':
     parser.add_argument("--nni_manager_nfs_mount_path", type=str)
     parser.add_argument("--container_nfs_mount_path", type=str)
     # args for kubeflow and frameworkController
-    parser.add_argument("--nfs_server", type=str)
     parser.add_argument("--nfs_path", type=str)
     parser.add_argument("--keyvault_vaultname", type=str)
     parser.add_argument("--keyvault_name", type=str)

test/nni_test/nnitest/run_tests.py

@@ -23,7 +23,7 @@ from utils import (CLEAR, EXPERIMENT_URL, GREEN, RED, REST_ENDPOINT,
 it_variables = {}
 
 
-def update_training_service_config(config, training_service, config_file_path):
+def update_training_service_config(config, training_service, config_file_path, nni_source_dir):
     it_ts_config = get_yml_content(os.path.join('config', 'training_service.yml'))
 
     # hack for kubeflow trial config
@@ -38,7 +38,7 @@ def update_training_service_config(config, training_service, config_file_path):
         config['trial'].pop('command')
         if 'gpuNum' in config['trial']:
             config['trial'].pop('gpuNum')
-    
+
     if training_service == 'adl':
         # hack for adl trial config, codeDir in adl mode refers to path in container
         containerCodeDir = config['trial']['codeDir']
@@ -52,6 +52,18 @@ def update_training_service_config(config, training_service, config_file_path):
             containerCodeDir = config['trial']['codeDir'].replace('../../../', '/')
         it_ts_config[training_service]['trial']['codeDir'] = containerCodeDir
         it_ts_config[training_service]['trial']['command'] = 'cd {0} && {1}'.format(containerCodeDir, config['trial']['command'])
+
+    if training_service == 'remote':
+        testcase_config = get_yml_content(nni_source_dir + config_file_path)
+        sharedStorage = testcase_config.get('sharedStorage')
+        if sharedStorage is None:
+            it_ts_config[training_service].pop('sharedStorage')
+        elif str(sharedStorage.get('storageType')).lower() == 'nfs':
+            it_ts_config[training_service].get('sharedStorage').pop('storageAccountKey')
+        elif str(sharedStorage.get('storageType')).lower() == 'azureblob':
+            it_ts_config[training_service].get('sharedStorage').pop('nfsServer')
+        else:
+            it_ts_config[training_service].pop('sharedStorage')
     
     if training_service == 'hybrid':
         it_ts_config = get_yml_content(os.path.join('config', 'training_service_v2.yml'))
@@ -75,7 +87,7 @@ def prepare_config_file(test_case_config, it_config, args):
     # apply training service config
     # user's gpuNum, logCollection config is overwritten by the config in training_service.yml
     # the hack for kubeflow should be applied at last step
-    update_training_service_config(test_yml_config, args.ts, test_case_config['configFile'])
+    update_training_service_config(test_yml_config, args.ts, test_case_config['configFile'], args.nni_source_dir)
 
     # generate temporary config yml file to launch experiment
     new_config_file = config_path + '.tmp'
@@ -238,6 +250,15 @@ def match_training_service(test_case_config, cur_training_service):
         return True
     return False
 
+def match_remoteConfig(test_case_config, nni_source_dir):
+    trainingservice_config = get_yml_content(os.path.join('config', 'training_service.yml'))
+    trainingservice_config_reuse_value = str(trainingservice_config['remote']['remoteConfig']['reuse']).lower()
+    testcase_config = get_yml_content(nni_source_dir + test_case_config['configFile'])
+    if testcase_config.get('remoteConfig') is not None:
+        if testcase_config['remoteConfig'].get('reuse') is not None:
+            return str(testcase_config['remoteConfig']['reuse']).lower() == trainingservice_config_reuse_value
+    return True
+
 
 def run(args):
     it_config = get_yml_content(args.config)
@@ -264,8 +285,13 @@ def run(args):
             print('skipped {}, training service {} not match [{}]'.format(
                 name, args.ts, test_case_config['trainingService']))
             continue
+
         # remote mode need more time to cleanup 
         if args.ts == 'remote' or args.ts == 'hybrid':
+            if args.ts == 'remote':
+                if not match_remoteConfig(test_case_config, args.nni_source_dir):
+                    print('skipped {}, remoteConfig not match.'.format(name))
+                    continue
             wait_for_port_available(8080, 240)
         else:
             wait_for_port_available(8080, 60)

test/nni_test/nnitest/validators.py

@@ -97,3 +97,17 @@ class NnicliValidator(ITValidator):
         print(exp.get_job_statistics())
         print(exp.get_experiment_status())
         print(exp.list_trial_jobs())
+
+class FileExistValidator(ITValidator):
+    def __call__(self, rest_endpoint, experiment_dir, nni_source_dir, **kwargs):
+        print(rest_endpoint)
+        exp_id = osp.split(experiment_dir)[-1]
+        rootpath = kwargs.get('rootpath')
+        
+        metrics = requests.get(METRICS_URL).json()
+        for metric in metrics:
+            trial_id = metric['trialJobId']
+            checkpath = osp.join(rootpath, 'nni', exp_id, 'trials', trial_id, 'nnioutput', 'checkingfile.txt')
+            print('Checking shared storage log exists on trial ',trial_id)
+            assert osp.exists(checkpath)
+

test/vso_tools/start_docker.py

@@ -25,8 +25,9 @@ container = sys.argv[2]
 password = sys.argv[3]
 
 run_command(f'docker build --build-arg NNI_RELEASE={version} -t nnidev/nni-nightly .')
-run_command(f'docker run -d -t -p {port}:22 --name {container} nnidev/nni-nightly')
+run_command(f'docker run --privileged -d -t -p {port}:22 --name {container} nnidev/nni-nightly')
 run_command(f'docker exec {container} useradd --create-home --password {password} nni')
 run_command(['docker', 'exec', container, 'bash', '-c', f'echo "nni:{password}" | chpasswd'])
+run_command(['docker', 'exec', container, 'bash', '-c', 'echo "nni ALL=(ALL:ALL) NOPASSWD:ALL" >> /etc/sudoers'])
 run_command(f'docker exec {container} service ssh start')
 set_variable('docker_port', port)

ts/nni_manager/training_service/reusable/shared_storages/azureblobStorageService.ts

@@ -34,13 +34,13 @@ fi
 id=$(lsb_release -i | cut -c16- | sed s/[[:space:]]//g)
 version=$(lsb_release -r | cut -c9- | sed s/[[:space:]]//g)
 
-if [ $id = "Ubuntu" ]
+if [ "$id" = "Ubuntu" ]
 then
     wget https://packages.microsoft.com/config/ubuntu/$version/packages-microsoft-prod.deb
-    sudo dpkg -i packages-microsoft-prod.deb
+    sudo DEBIAN_FRONTEND=noninteractive dpkg -i packages-microsoft-prod.deb
     sudo apt-get update
     sudo apt-get install -y blobfuse fuse
-elif [ $id = "CentOS" ] || [ $id = "RHEL" ]
+elif [ "$id" = "CentOS" ] || [ "$id" = "RHEL" ]
 then
     sudo rpm -Uvh https://packages.microsoft.com/config/rhel/$(echo $version | cut -c1)/packages-microsoft-prod.rpm
     sudo yum install -y blobfuse fuse