Merge pull request #116 from Microsoft/master

merge master

test/utils.py

@@ -27,6 +27,16 @@ import yaml
 
 EXPERIMENT_DONE_SIGNAL = '"Experiment done"'
 
+GREEN = '\33[32m'
+RED = '\33[31m'
+CLEAR = '\33[0m'
+
+REST_ENDPOINT = 'http://localhost:8080/api/v1/nni'
+EXPERIMENT_URL = REST_ENDPOINT + '/experiment'
+STATUS_URL = REST_ENDPOINT + '/check-status'
+TRIAL_JOBS_URL = REST_ENDPOINT + '/trial-jobs'
+METRICS_URL = REST_ENDPOINT + '/metric-data'
+
 def read_last_line(file_name):
     '''read last line of a file and return None if file not found'''
     try:
@@ -73,10 +83,38 @@ def fetch_nni_log_path(experiment_url):
 
     return nnimanager_log_path
 
-def check_experiment_status(nnimanager_log_path):
+def is_experiment_done(nnimanager_log_path):
     '''check if the experiment is done successfully'''
     assert os.path.exists(nnimanager_log_path), 'Experiment starts failed'
     cmds = ['cat', nnimanager_log_path, '|', 'grep', EXPERIMENT_DONE_SIGNAL]
     completed_process = subprocess.run(' '.join(cmds), shell=True)
 
     return completed_process.returncode == 0
+
+def get_experiment_status(status_url):
+    nni_status = requests.get(status_url).json()
+    #print(nni_status)
+    return nni_status['status']
+
+def get_succeeded_trial_num(trial_jobs_url):
+    trial_jobs = requests.get(trial_jobs_url).json()
+    print(trial_jobs)
+    num_succeed = 0
+    for trial_job in trial_jobs:
+        if trial_job['status'] in ['SUCCEEDED', 'EARLY_STOPPED']:
+            num_succeed += 1
+    print('num_succeed:', num_succeed)
+    return num_succeed
+
+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]
+            subprocess.run(['cat', stderr_path])
+
+def parse_max_duration_time(max_exec_duration):
+    unit = max_exec_duration[-1]
+    time = max_exec_duration[:-1]
+    units_dict = {'s':1, 'm':60, 'h':3600, 'd':86400}
+    return int(time) * units_dict[unit]

tools/nni_cmd/config_schema.py

@@ -216,12 +216,14 @@ frameworkcontroller_trial_schema = {
 frameworkcontroller_config_schema = {
     'frameworkcontrollerConfig':Or({
         Optional('storage'): Or('nfs', 'azureStorage'),
+        Optional('serviceAccountName'): str,
         'nfs': {
             'server': str,
             'path': str
         }
     },{
         Optional('storage'): Or('nfs', 'azureStorage'),
+        Optional('serviceAccountName'): str,
         'keyVault': {
             'vaultName': Regex('([0-9]|[a-z]|[A-Z]|-){1,127}'),
             'name': Regex('([0-9]|[a-z]|[A-Z]|-){1,127}')

tools/nni_cmd/launcher_utils.py

@@ -35,18 +35,18 @@ def parse_relative_path(root_path, experiment_config, key):
         print_normal('expand %s: %s to %s ' % (key, experiment_config[key], absolute_path))
         experiment_config[key] = absolute_path
 
-def parse_time(experiment_config):
-    '''Parse time format'''
-    unit = experiment_config['maxExecDuration'][-1]
+def parse_time(time):
+    '''Change the time to seconds'''
+    unit = time[-1]
     if unit not in ['s', 'm', 'h', 'd']:
         print_error('the unit of time could only from {s, m, h, d}')
         exit(1)
-    time = experiment_config['maxExecDuration'][:-1]
+    time = time[:-1]
     if not time.isdigit():
         print_error('time format error!')
         exit(1)
     parse_dict = {'s':1, 'm':60, 'h':3600, 'd':86400}
-    experiment_config['maxExecDuration'] = int(time) * parse_dict[unit]
+    return int(time) * parse_dict[unit]
 
 def parse_path(experiment_config, config_path):
     '''Parse path in config file'''
@@ -216,7 +216,7 @@ def validate_all_content(experiment_config, config_path):
     '''Validate whether experiment_config is valid'''
     parse_path(experiment_config, config_path)
     validate_common_content(experiment_config)
-    parse_time(experiment_config)
+    experiment_config['maxExecDuration'] = parse_time(experiment_config['maxExecDuration'])
     if experiment_config.get('advisor'):
         parse_advisor_content(experiment_config)
         validate_annotation_content(experiment_config, 'advisor', 'builtinAdvisorName')

tools/nni_cmd/nnictl.py

@@ -73,7 +73,7 @@ def parse_args():
     parser_updater_concurrency.set_defaults(func=update_concurrency)
     parser_updater_duration = parser_updater_subparsers.add_parser('duration', help='update duration')
     parser_updater_duration.add_argument('id', nargs='?', help='the id of experiment')
-    parser_updater_duration.add_argument('--value', '-v', required=True)
+    parser_updater_duration.add_argument('--value', '-v', required=True, help='the unit of time should in {\'s\', \'m\', \'h\', \'d\'}')
     parser_updater_duration.set_defaults(func=update_duration)
     parser_updater_trialnum = parser_updater_subparsers.add_parser('trialnum', help='update maxtrialnum')
     parser_updater_trialnum.add_argument('--id', '-i', dest='id', help='the id of experiment')

tools/nni_cmd/updater.py

@@ -26,6 +26,7 @@ from .url_utils import experiment_url
 from .config_utils import Config
 from .common_utils import get_json_content
 from .nnictl_utils import check_experiment_id, get_experiment_port, get_config_filename
+from .launcher_utils import parse_time
 
 def validate_digit(value, start, end):
     '''validate if a digit is valid'''
@@ -92,7 +93,8 @@ def update_concurrency(args):
             print('ERROR: update %s failed!' % 'concurrency')
 
 def update_duration(args):
-    validate_digit(args.value, 1, 999999999)
+    #parse time, change time unit to seconds 
+    args.value = parse_time(args.value)
     args.port = get_experiment_port(args)
     if args.port is not None:
         if update_experiment_profile(args, 'maxExecDuration', int(args.value)):