Refactor integration test (step 2) - AML authentication and debuggability (#4893)

ts/nni_manager/config/aml/amlUtil.py

@@ -5,11 +5,14 @@ import os
 import sys
 import time
 import json
+import warnings
 from argparse import ArgumentParser
-from azureml.core import Experiment, RunConfiguration, ScriptRunConfig
+from azureml.core import Experiment, RunConfiguration, ScriptRunConfig, Workspace
+from azureml.core.authentication import (
+    AzureCliAuthentication, InteractiveLoginAuthentication, AuthenticationException
+)
 from azureml.core.compute import ComputeTarget
 from azureml.core.run import RUNNING_STATES, RunStatus, Run
-from azureml.core import Workspace
 from azureml.core.conda_dependencies import CondaDependencies
 
 if __name__ == "__main__":
@@ -24,7 +27,18 @@ if __name__ == "__main__":
     parser.add_argument('--script_name', help='script name')
     args = parser.parse_args()
 
-    ws = Workspace(args.subscription_id, args.resource_group, args.workspace_name)
+    try:
+        auth = AzureCliAuthentication()
+        auth.get_token()
+    except AuthenticationException as e:
+        warnings.warn(
+            f'Azure-cli authentication failed: {e}',
+            RuntimeWarning
+        )
+        warnings.warn('Falling back to interactive authentication.', RuntimeWarning)
+        auth = InteractiveLoginAuthentication()
+
+    ws = Workspace(args.subscription_id, args.resource_group, args.workspace_name, auth=auth)
     compute_target = ComputeTarget(workspace=ws, name=args.compute_target)
     experiment = Experiment(ws, args.experiment_name)
     run_config = RunConfiguration()

ts/nni_manager/core/nnimanager.ts

@@ -728,6 +728,7 @@ class NNIManager implements Manager {
                 throw NNIError.FromError(err, 'Dispatcher error: ');
             }),
             this.trainingService.run().catch((err: Error) => {
+                // FIXME: The error handling here could crash when err is undefined.
                 throw NNIError.FromError(err, 'Training service error: ');
             }),
             this.manageTrials().catch((err: Error) => {
@@ -870,10 +871,15 @@ class NNIManager implements Manager {
     }
 
     private logError(err: Error): void {
+        if (err !== undefined) {
+            // FIXME: I don't know why, but in some cases err could be undefined.
             if (err.stack !== undefined) {
                 this.log.error(err.stack);
-        }
                 this.status.errors.push(err.message);
+            } else {
+                this.status.errors.push(`Undefined error, stack: ${new Error().stack}`);
+            }
+        }
         this.setEndtime();
         this.setStatus('ERROR');
     }

ts/nni_manager/training_service/reusable/aml/amlClient.ts

@@ -135,12 +135,18 @@ export class AMLClient {
     
     // Monitor error information in aml python shell client
     private monitorError(pythonShellClient: PythonShell, deferred: Deferred<any>): void {
-        pythonShellClient.on('error', function (error: any) {
-            deferred.reject(error);
+        pythonShellClient.on('stderr', function (chunk: any) {
+            // FIXME: The error will only appear in console.
+            // Still need to find a way to put them into logs.
+            console.error(`Python process stderr: ${chunk}`);
         });
-        pythonShellClient.on('close', function (error: any) {
+        pythonShellClient.on('error', function (error: Error) {
+            console.error(`Python process fires error: ${error}`);
             deferred.reject(error);
         });
+        pythonShellClient.on('close', function () {
+            deferred.reject(new Error('AML client Python process unknown error.'));
+        });
     }
     
     // Parse command content, command format is {head}:{content}

ts/nni_manager/training_service/reusable/environments/amlEnvironmentService.ts

@@ -109,6 +109,7 @@ export class AMLEnvironmentService extends EnvironmentService {
             'nni_script.py',
             environmentLocalTempFolder
         );
+        this.log.debug('aml: before amlClient submit');
         amlEnvironment.id = await amlClient.submit();
         this.log.debug('aml: before getTrackingUrl');
         amlEnvironment.trackingUrl = await amlClient.getTrackingUrl();