Windows pr pipeline (microsoft#1682)

azure-pipelines.yml

@@ -13,8 +13,8 @@ jobs:
   - script: |
       python3 -m pip install torch==0.4.1 --user
       python3 -m pip install torchvision==0.2.1 --user
-      python3 -m pip install tensorflow==1.12.0 --user
-    displayName: 'Install dependencies for integration'
+      python3 -m pip install tensorflow==1.13.1 --user
+    displayName: 'Install dependencies'
   - script: |
       source install.sh
     displayName: 'Install nni toolkit via source code'
@@ -59,7 +59,7 @@ jobs:
       python3 -m pip install torch==0.4.1 --user
       python3 -m pip install torchvision==0.2.1 --user
       python3 -m pip install tensorflow==1.13.1 --user
-    displayName: 'Install dependencies for integration'
+    displayName: 'Install dependencies'
   - script: |
       source install.sh
     displayName: 'Install nni toolkit via source code'
@@ -79,3 +79,43 @@ jobs:
       cd test
       PATH=$HOME/Library/Python/3.7/bin:$PATH python3 cli_test.py
     displayName: 'nnicli test'
+
+- job: 'basic_test_pr_Windows'
+  pool:
+    vmImage: 'vs2017-win2016'
+  strategy:
+    matrix:
+      Python36:
+        PYTHON_VERSION: '3.6'
+
+  steps:
+  - script: |
+      powershell.exe -file install.ps1
+    displayName: 'Install nni toolkit via source code'
+  - script: |
+      python -m pip install scikit-learn==0.20.0 --user
+      python -m pip install keras==2.1.6 --user
+      python -m pip install https://download.pytorch.org/whl/cu90/torch-0.4.1-cp36-cp36m-win_amd64.whl --user
+      python -m pip install torchvision --user
+      python -m pip install tensorflow==1.13.1 --user
+    displayName: 'Install dependencies'
+  - script: |
+      cd test
+      powershell.exe -file unittest.ps1
+    displayName: 'unit test'
+  - script: |
+      cd test
+      python naive_test.py
+    displayName: 'Naive test'
+  - script: |
+      cd test
+      python tuner_test.py
+    displayName: 'Built-in tuners / assessors tests'
+  - script: |
+      cd test
+      python metrics_test.py
+    displayName: 'Trial job metrics test'
+  - script: |
+      cd test
+      PATH=$HOME/.local/bin:$PATH python3 cli_test.py
+    displayName: 'nnicli test'

test/metrics_test.py

@@ -18,18 +18,23 @@
 # DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 # OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
+import sys
+import os.path as osp
 import subprocess
 import time
 import traceback
 import json
 import requests
 
-from utils import get_experiment_status, get_yml_content, parse_max_duration_time, get_succeeded_trial_num
+from utils import get_experiment_status, get_yml_content, parse_max_duration_time, get_succeeded_trial_num, print_stderr
 from utils import GREEN, RED, CLEAR, STATUS_URL, TRIAL_JOBS_URL, METRICS_URL
 
 def run_test():
     '''run metrics test'''
-    config_file = 'metrics_test/metrics.test.yml'
+    if sys.platform == 'win32':
+        config_file = osp.join('metrics_test', 'metrics_win32.test.yml')
+    else:
+        config_file = osp.join('metrics_test', 'metrics.test.yml')
 
     print('Testing %s...' % config_file)
     proc = subprocess.run(['nnictl', 'create', '--config', config_file])
@@ -44,14 +49,15 @@ def run_test():
         #print('experiment status:', status)
         if status == 'DONE':
             num_succeeded = get_succeeded_trial_num(TRIAL_JOBS_URL)
+            print_stderr(TRIAL_JOBS_URL)
             assert num_succeeded == max_trial_num, 'only %d succeeded trial jobs, there should be %d' % (num_succeeded, max_trial_num)
             check_metrics()
             break
 
     assert status == 'DONE', 'Failed to finish in maxExecDuration'
 
 def check_metrics():
-    with open('metrics_test/expected_metrics.json', 'r') as f:
+    with open(osp.join('metrics_test', 'expected_metrics.json'), 'r') as f:
         expected_metrics = json.load(f)
     print(expected_metrics)
     metrics = requests.get(METRICS_URL).json()

test/metrics_test/metrics_win32.test.yml

@@ -0,0 +1,20 @@
+authorName: nni
+experimentName: default_test
+maxExecDuration: 3m
+maxTrialNum: 1
+trialConcurrency: 1
+searchSpacePath: ./search_space.json
+
+tuner:
+  builtinTunerName: Random
+
+trial:
+  codeDir: .
+  command: python trial.py
+  gpuNum: 0
+
+useAnnotation: false
+multiPhase: false
+multiThread: false
+
+trainingServicePlatform: local

test/naive_test.py

@@ -18,6 +18,8 @@
 # DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 # OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
+import sys
+import os.path as osp
 import json
 import subprocess
 import sys
@@ -30,10 +32,14 @@
 def naive_test():
     '''run naive integration test'''
     to_remove = ['tuner_search_space.json', 'tuner_result.txt', 'assessor_result.txt']
-    to_remove = list(map(lambda file: 'naive_test/' + file, to_remove))
+    to_remove = list(map(lambda file: osp.join('naive_test', file), to_remove))
     remove_files(to_remove)
 
-    proc = subprocess.run(['nnictl', 'create', '--config', 'naive_test/local.yml'])
+    if sys.platform == 'win32':
+        config_file = 'local_win32.yml'
+    else:
+        config_file = 'local.yml'
+    proc = subprocess.run(['nnictl', 'create', '--config', osp.join('naive_test' , config_file)])
     assert proc.returncode == 0, '`nnictl create` failed with code %d' % proc.returncode
 
     print('Spawning trials...')
@@ -44,8 +50,8 @@ def naive_test():
     for _ in range(120):
         time.sleep(1)
 
-        tuner_status = read_last_line('naive_test/tuner_result.txt')
-        assessor_status = read_last_line('naive_test/assessor_result.txt')
+        tuner_status = read_last_line(osp.join('naive_test', 'tuner_result.txt'))
+        assessor_status = read_last_line(osp.join('naive_test', 'assessor_result.txt'))
         experiment_status = is_experiment_done(nnimanager_log_path)
 
         assert tuner_status != 'ERROR', 'Tuner exited with error'
@@ -55,7 +61,7 @@ def naive_test():
             break
 
         if tuner_status is not None:
-            for line in open('naive_test/tuner_result.txt'):
+            for line in open(osp.join('naive_test', 'tuner_result.txt')):
                 if line.strip() == 'ERROR':
                     break
                 trial = int(line.split(' ')[0])
@@ -65,18 +71,20 @@ def naive_test():
 
     assert experiment_status, 'Failed to finish in 2 min'
 
-    ss1 = json.load(open('naive_test/search_space.json'))
-    ss2 = json.load(open('naive_test/tuner_search_space.json'))
+    ss1 = json.load(open(osp.join('naive_test', 'search_space.json')))
+    ss2 = json.load(open(osp.join('naive_test', 'tuner_search_space.json')))
     assert ss1 == ss2, 'Tuner got wrong search space'
 
-    tuner_result = set(open('naive_test/tuner_result.txt'))
-    expected = set(open('naive_test/expected_tuner_result.txt'))
+    tuner_result = set(open(osp.join('naive_test', 'tuner_result.txt')))
+    expected = set(open(osp.join('naive_test', 'expected_tuner_result.txt')))
     # Trials may complete before NNI gets assessor's result,
     # so it is possible to have more final result than expected
+    print('Tuner result:', tuner_result)
+    print('Expected tuner result:', expected)
     assert tuner_result.issuperset(expected), 'Bad tuner result'
 
-    assessor_result = set(open('naive_test/assessor_result.txt'))
-    expected = set(open('naive_test/expected_assessor_result.txt'))
+    assessor_result = set(open(osp.join('naive_test', 'assessor_result.txt')))
+    expected = set(open(osp.join('naive_test', 'expected_assessor_result.txt')))
     assert assessor_result == expected, 'Bad assessor result'
 
     subprocess.run(['nnictl', 'stop'])
@@ -85,10 +93,10 @@ def naive_test():
 def stop_experiment_test():
     '''Test `nnictl stop` command, including `nnictl stop exp_id` and `nnictl stop all`.
     Simple `nnictl stop` is not tested here since it is used in all other test code'''
-    subprocess.run(['nnictl', 'create', '--config', 'tuner_test/local.yml', '--port', '8080'], check=True)
-    subprocess.run(['nnictl', 'create', '--config', 'tuner_test/local.yml', '--port', '8888'], check=True)
-    subprocess.run(['nnictl', 'create', '--config', 'tuner_test/local.yml', '--port', '8989'], check=True)
-    subprocess.run(['nnictl', 'create', '--config', 'tuner_test/local.yml', '--port', '8990'], check=True)
+    subprocess.run(['nnictl', 'create', '--config', osp.join('tuner_test', 'local.yml'), '--port', '8080'], check=True)
+    subprocess.run(['nnictl', 'create', '--config', osp.join('tuner_test', 'local.yml'), '--port', '8888'], check=True)
+    subprocess.run(['nnictl', 'create', '--config', osp.join('tuner_test', 'local.yml'), '--port', '8989'], check=True)
+    subprocess.run(['nnictl', 'create', '--config', osp.join('tuner_test', 'local.yml'), '--port', '8990'], check=True)
 
     # test cmd 'nnictl stop id`
     experiment_id = get_experiment_id(EXPERIMENT_URL)

test/naive_test/local_win32.yml

@@ -0,0 +1,26 @@
+authorName: nni
+experimentName: naive
+trialConcurrency: 3
+maxExecDuration: 1h
+maxTrialNum: 10
+#choice: local, remote
+trainingServicePlatform: local
+searchSpacePath: search_space.json
+#choice: true, false
+useAnnotation: false
+tuner:
+    codeDir: .
+    classFileName: naive_tuner.py
+    className: NaiveTuner
+    classArgs:
+        optimize_mode: maximize
+assessor:
+    codeDir: .
+    classFileName: naive_assessor.py
+    className: NaiveAssessor
+    classArgs:
+        optimize_mode: maximize
+trial:
+    command: python naive_trial.py
+    codeDir: .
+    gpuNum: 0

test/tuner_test.py

@@ -18,6 +18,8 @@
 # DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 # OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
+import sys
+import os.path as osp
 import subprocess
 import sys
 import time
@@ -30,9 +32,16 @@
 ASSESSOR_LIST = ['Medianstop']
 
 
+def get_config_file_path():
+    if sys.platform == 'win32':
+        config_file = osp.join('tuner_test', 'local_win32.yml')
+    else:
+        config_file = osp.join('tuner_test', 'local.yml')
+    return config_file
+
 def switch(dispatch_type, dispatch_name):
     '''Change dispatch in config.yml'''
-    config_path = 'tuner_test/local.yml'
+    config_path = get_config_file_path()
     experiment_config = get_yml_content(config_path)
     if dispatch_name in ['GridSearch', 'BatchTuner', 'Random']:
         experiment_config[dispatch_type.lower()] = {
@@ -56,7 +65,7 @@ def test_builtin_dispatcher(dispatch_type, dispatch_name):
     switch(dispatch_type, dispatch_name)
 
     print('Testing %s...' % dispatch_name)
-    proc = subprocess.run(['nnictl', 'create', '--config', 'tuner_test/local.yml'])
+    proc = subprocess.run(['nnictl', 'create', '--config', get_config_file_path()])
     assert proc.returncode == 0, '`nnictl create` failed with code %d' % proc.returncode
 
     nnimanager_log_path = get_nni_log_path(EXPERIMENT_URL)

test/tuner_test/local_win32.yml

@@ -0,0 +1,20 @@
+assessor:
+  builtinAssessorName: Medianstop
+  classArgs:
+    optimize_mode: maximize
+authorName: nni
+experimentName: test_sdk
+maxExecDuration: 1h
+maxTrialNum: 2
+searchSpacePath: search_space.json
+trainingServicePlatform: local
+trial:
+  codeDir: .
+  command: python naive_trial.py
+  gpuNum: 0
+trialConcurrency: 2
+tuner:
+  builtinTunerName: Evolution
+  classArgs:
+    optimize_mode: maximize
+useAnnotation: false

test/unittest.ps1

@@ -1,5 +1,5 @@
 $CWD = $PWD
-
+$ErrorActionPreference = "Stop"
 # -------------For python unittest-------------
 
 ## ------Run annotation test------

test/utils.py

@@ -20,7 +20,6 @@
 
 import contextlib
 import collections
-import json
 import os
 import socket
 import sys
@@ -29,7 +28,7 @@
 import time
 import ruamel.yaml as yaml
 
-EXPERIMENT_DONE_SIGNAL = '"Experiment done"'
+EXPERIMENT_DONE_SIGNAL = 'Experiment done'
 
 GREEN = '\33[32m'
 RED = '\33[31m'
@@ -93,13 +92,11 @@ def get_nni_log_path(experiment_url):
 def is_experiment_done(nnimanager_log_path):
     '''check if the experiment is done successfully'''
     assert os.path.exists(nnimanager_log_path), 'Experiment starts failed'
-    if sys.platform == "win32":
-        cmds = ['type', nnimanager_log_path, '|', 'find', EXPERIMENT_DONE_SIGNAL]
-    else:
-        cmds = ['cat', nnimanager_log_path, '|', 'grep', EXPERIMENT_DONE_SIGNAL]
-    completed_process = subprocess.run(' '.join(cmds), shell=True)
-
-    return completed_process.returncode == 0
+
+    with open(nnimanager_log_path, 'r') as f:
+        log_content = f.read()
+
+    return EXPERIMENT_DONE_SIGNAL in log_content
 
 def get_experiment_status(status_url):
     nni_status = requests.get(status_url).json()
@@ -119,10 +116,12 @@ def print_stderr(trial_jobs_url):
     trial_jobs = requests.get(trial_jobs_url).json()
     for trial_job in trial_jobs:
         if trial_job['status'] == 'FAILED':
-            stderr_path = trial_job['stderrPath'].split(':')[-1]
             if sys.platform == "win32":
+                p = trial_job['stderrPath'].split(':')
+                stderr_path = ':'.join([p[-2], p[-1]])
                 subprocess.run(['type', stderr_path], shell=True)
             else:
+                stderr_path = trial_job['stderrPath'].split(':')[-1]
                 subprocess.run(['cat', stderr_path])
 
 def parse_max_duration_time(max_exec_duration):