Add kubeflow and frameworkcontroller hybrid example (#4343)

docs/en_US/TrainingService/HybridMode.rst

@@ -6,7 +6,7 @@ Run NNI on hybrid mode means that NNI will run trials jobs in multiple kinds of
 Setup environment
 -----------------
 
-NNI has supported `local <./LocalMode.rst>`__\ , `remote <./RemoteMachineMode.rst>`__\ , `PAI <./PaiMode.rst>`__\ , and `AML <./AMLMode.rst>`__ for hybrid training service. Before starting an experiment using these mode, users should setup the corresponding environment for the platforms. More details about the environment setup could be found in the corresponding docs.
+NNI has supported `local <./LocalMode.rst>`__\ , `remote <./RemoteMachineMode.rst>`__\ , `PAI <./PaiMode.rst>`__\ , `AML <./AMLMode.rst>`__,  `Kubeflow <./KubeflowMode.rst>`__\ , `FrameworkController <./FrameworkControllerMode.rst>`__\ ,for hybrid training service. Before starting an experiment using these mode, users should setup the corresponding environment for the platforms. More details about the environment setup could be found in the corresponding docs.
 
 Run an experiment
 -----------------
@@ -36,4 +36,4 @@ Use ``examples/trials/mnist-tfv1`` as an example. The NNI config YAML file's con
       - platform: local
 
 To use hybrid training services, users should set training service configurations as a list in `trainingService` field.  
-Currently, hybrid support setting `local`, `remote`, `pai` and `aml` training services.
+Currently, hybrid support setting `local`, `remote`, `pai`, `aml`, `kubeflow` and `frameworkcontroller` training services.

examples/trials/mnist-tfv2/config_hybrid.yml

@@ -21,3 +21,41 @@ trainingService:
     resourceGroup: ${your resource group}
     workspaceName: ${your workspace name}
     computeTarget: ${your compute target}
+  - platform: kubeflow
+    reuseMode: true
+    worker:
+      command:
+      code_directory: 
+      dockerImage: msranni/nni
+      cpuNumber: 
+      gpuNumber: 
+      memorySize: 
+      replicas: 
+    operator: tf-operator
+    storage:
+      storageType: 
+      azureAccount: 
+      azureShare: 
+      keyVaultName: 
+      keyVaultKey: 
+    apiVersion: v1
+  - platform: frameworkcontroller
+    reuseMode: true
+    serviceAccountName: 
+    taskRoles:
+      - name: worker
+        dockerImage: 'msranni/nni:latest'
+        taskNumber: 
+        command: 
+        gpuNumber: 
+        cpuNumber: 
+        memorySize: 
+        framework_attempt_completion_policy:
+          min_failed_task_count: 1
+          minSucceedTaskCount: 1
+    storage:
+      storageType: 
+      azureAccount: 
+      azureShare: 
+      keyVaultName: 
+      keyVaultKey: 

pipelines/integration-test-frameworkcontroller.yml

@@ -56,7 +56,7 @@ jobs:
       cd test
       python3 nni_test/nnitest/generate_ts_config.py \
           --ts frameworkcontroller \
-          --keyvault_vaultname $(keyvault_vaultname) \
+          --keyvault_vaultname  $(keyvault_vaultname) \
           --keyvault_name $(keyvault_name) \
           --azs_account $(azs_account) \
           --azs_share $(azs_share) \
@@ -64,5 +64,5 @@ jobs:
           --nni_manager_ip $(manager_ip) \
           --reuse_mode True \
           --config_version v2
-      python3 nni_test/nnitest/run_tests.py --config config/integration_tests.yml --ts frameworkcontroller --reuse_mode True --exclude multi-phase,multi-thread
+      python3 nni_test/nnitest/run_tests.py --config config/integration_tests_config_v2.yml --ts frameworkcontroller --reuse_mode True --exclude multi-phase,multi-thread
     displayName: Integration test (reuse mode)

pipelines/integration-test-kubeflow.yml

@@ -68,7 +68,7 @@ jobs:
       az login --service-principal -u $(client_id) -p $(client_secret) --tenant $(tenant_id)
       python3 nni_test/nnitest/generate_ts_config.py \
           --ts kubeflow \
-          --keyvault_vaultname $(keyvault_vaultname) \``
+          --keyvault_vaultname $(keyvault_vaultname) \
           --keyvault_name $(keyvault_name) \
           --azs_account $(azs_account) \
           --azs_share $(azs_share) \
@@ -76,5 +76,5 @@ jobs:
           --nni_manager_ip $(manager_ip) \
           --reuse_mode True \
           --config_version v2
-      python3 nni_test/nnitest/run_tests.py --config config/integration_tests.yml --ts kubeflow --reuse_mode True --exclude multi-phase,multi-thread
+      python3 nni_test/nnitest/run_tests.py --config config/integration_tests_config_v2.yml --ts kubeflow --reuse_mode True --exclude multi-phase,multi-thread
     displayName: Integration test (reuse mode)

test/config/training_service_v2.yml

@@ -13,7 +13,7 @@ hybrid:
     workspaceName:
     computeTarget:
 kubeflow:
-  trialGpuNumber: 0
+  trialGpuNumber: 1
   trialConcurrency: 2
   maxTrialNumber: 2
   nniManagerIp:
@@ -37,7 +37,7 @@ kubeflow:
       keyVaultKey:
     apiVersion: v1
 frameworkcontroller:
-  trialGpuNumber: 0
+  trialGpuNumber: 1
   trialConcurrency: 2
   maxTrialNumber: 2
   nniManagerIp:

test/nni_test/nnitest/generate_ts_config.py

@@ -55,8 +55,8 @@ def update_training_service_config(args):
         config[args.ts]['trainingService']['worker']['dockerImage'] = args.nni_docker_image
         config[args.ts]['trainingService']['storage']['azureAccount'] = args.azs_account
         config[args.ts]['trainingService']['storage']['azureShare'] = args.azs_share
-        config[args.ts]['trainingService']['storage']['keyVaultName'] = args.keyvault_name
-        config[args.ts]['trainingService']['storage']['keyVaultKey'] = args.keyvault_vaultname
+        config[args.ts]['trainingService']['storage']['keyVaultName'] = args.keyvault_vaultname
+        config[args.ts]['trainingService']['storage']['keyVaultKey'] = args.keyvault_name
         config[args.ts]['nni_manager_ip'] = args.nni_manager_ip
         dump_yml_content(TRAINING_SERVICE_FILE_V2, config)
     elif args.ts == 'frameworkcontroller' and args.reuse_mode == 'False':
@@ -79,8 +79,8 @@ def update_training_service_config(args):
         config[args.ts]['trainingService']['taskRoles'][0]['dockerImage'] = args.nni_docker_image
         config[args.ts]['trainingService']['storage']['azureAccount'] = args.azs_account
         config[args.ts]['trainingService']['storage']['azureShare'] = args.azs_share
-        config[args.ts]['trainingService']['storage']['keyVaultName'] = args.keyvault_name
-        config[args.ts]['trainingService']['storage']['keyVaultKey'] = args.keyvault_vaultname
+        config[args.ts]['trainingService']['storage']['keyVaultName'] = args.keyvault_vaultname
+        config[args.ts]['trainingService']['storage']['keyVaultKey'] = args.keyvault_name
         config[args.ts]['nni_manager_ip'] = args.nni_manager_ip
         dump_yml_content(TRAINING_SERVICE_FILE_V2, config)
     elif args.ts == 'remote':

test/nni_test/nnitest/run_tests.py

@@ -25,7 +25,6 @@ it_variables = {}
 
 def update_training_service_config(config, training_service, config_file_path, nni_source_dir, reuse_mode='False'):
     it_ts_config = get_yml_content(os.path.join('config', 'training_service.yml'))
-
     # hack for kubeflow trial config
     if training_service == 'kubeflow' and reuse_mode == 'False':
         it_ts_config[training_service]['trial']['worker']['command'] = config['trial']['command']
@@ -34,16 +33,18 @@ def update_training_service_config(config, training_service, config_file_path, n
             config['trial'].pop('gpuNum')
     elif training_service == 'kubeflow' and reuse_mode == 'True':
         it_ts_config = get_yml_content(os.path.join('config', 'training_service_v2.yml'))
-        it_ts_config['trainingService']['worker']['command'] = config['trialCommand']
+        print(it_ts_config)
+        it_ts_config[training_service]['trainingService']['worker']['command'] = config['trialCommand']
+        it_ts_config[training_service]['trainingService']['worker']['code_directory'] = config['trialCodeDirectory']
 
-    if training_service == 'frameworkcontroller'  and reuse_mode == 'False':
+    if training_service == 'frameworkcontroller' and reuse_mode == 'False':
         it_ts_config[training_service]['trial']['taskRoles'][0]['command'] = config['trial']['command']
         config['trial'].pop('command')
         if 'gpuNum' in config['trial']:
             config['trial'].pop('gpuNum')
-    elif training_service == 'frameworkcontroller'  and reuse_mode == 'True':
+    elif training_service == 'frameworkcontroller' and reuse_mode == 'True':
         it_ts_config = get_yml_content(os.path.join('config', 'training_service_v2.yml'))
-        it_ts_config['trainingService']['taskRoles'][0]['command'] = config['trialCommand']
+        it_ts_config[training_service]['trainingService']['taskRoles'][0]['command'] = config['trialCommand']
 
     if training_service == 'adl':
         # hack for adl trial config, codeDir in adl mode refers to path in container
@@ -74,7 +75,7 @@ def update_training_service_config(config, training_service, config_file_path, n
     
     if training_service == 'hybrid':
         it_ts_config = get_yml_content(os.path.join('config', 'training_service_v2.yml'))
-    else:
+    elif reuse_mode != 'True':
         deep_update(config, it_ts_config['all'])
     deep_update(config, it_ts_config[training_service])