Update hpo with master (aws#17)

Author: laurenyu

CHANGELOG.rst

@@ -11,6 +11,7 @@ CHANGELOG
 * bug-fix: tensorflow-serving-api: SageMaker does not conflict with tensorflow-serving-api module version
 * feature: Local Mode: add support for local training data using file://
 * feature: Updated TensorFlow Serving api protobuf files
+* bug-fix: No longer poll for logs from stopped training jobs
 
 1.2.4
 =====

src/sagemaker/estimator.py

@@ -25,6 +25,7 @@
 from sagemaker.model import Model
 from sagemaker.model import (SCRIPT_PARAM_NAME, DIR_PARAM_NAME, CLOUDWATCH_METRICS_PARAM_NAME,
                              CONTAINER_LOG_LEVEL_PARAM_NAME, JOB_NAME_PARAM_NAME, SAGEMAKER_REGION_PARAM_NAME)
+
 from sagemaker.predictor import RealTimePredictor
 from sagemaker.session import Session
 from sagemaker.session import s3_input

src/sagemaker/session.py

@@ -480,7 +480,7 @@ def _check_job_status(self, job, desc):
         """
         status = desc['TrainingJobStatus']
 
-        if status != 'Completed':
+        if status != 'Completed' and status != 'Stopped':
             reason = desc.get('FailureReason', '(No reason provided)')
             raise ValueError('Error training {}: {} Reason: {}'.format(job, status, reason))
 
@@ -666,7 +666,7 @@ def logs_for_job(self, job_name, wait=False, poll=10):  # noqa: C901 - suppress
         client = self.boto_session.client('logs', config=config)
         log_group = '/aws/sagemaker/TrainingJobs'
 
-        job_already_completed = True if status == 'Completed' or status == 'Failed' else False
+        job_already_completed = True if status == 'Completed' or status == 'Failed' or status == 'Stopped' else False
 
         state = LogState.TAILING if wait and not job_already_completed else LogState.COMPLETE
         dot = False
@@ -738,7 +738,7 @@ def logs_for_job(self, job_name, wait=False, poll=10):  # noqa: C901 - suppress
 
                 status = description['TrainingJobStatus']
 
-                if status == 'Completed' or status == 'Failed':
+                if status == 'Completed' or status == 'Failed' or status == 'Stopped':
                     state = LogState.JOB_COMPLETE
 
         if wait:

tests/unit/test_estimator.py

@@ -402,6 +402,11 @@ def test_wait_with_logs(sagemaker_session):
     assert not sagemaker_session.wait_for_job.called
 
 
+def test_unsupported_type_in_dict():
+    with pytest.raises(ValueError):
+        _TrainingJob._format_inputs_to_input_config({'a': 66})
+
+
 #################################################################################
 # Tests for the generic Estimator class
 

tests/unit/test_session.py

@@ -185,6 +185,10 @@ def test_s3_input_all_arguments():
     {'TrainingStartTime': datetime.datetime(2018, 2, 17, 7, 15, 0, 103000)})
 COMPLETED_DESCRIBE_JOB_RESULT.update(
     {'TrainingEndTime': datetime.datetime(2018, 2, 17, 7, 19, 34, 953000)})
+
+STOPPED_DESCRIBE_JOB_RESULT = dict(COMPLETED_DESCRIBE_JOB_RESULT)
+STOPPED_DESCRIBE_JOB_RESULT.update({'TrainingJobStatus': 'Stopped'})
+
 IN_PROGRESS_DESCRIBE_JOB_RESULT = dict(DEFAULT_EXPECTED_TRAIN_JOB_ARGS)
 IN_PROGRESS_DESCRIBE_JOB_RESULT.update({'TrainingJobStatus': 'InProgress'})
 
@@ -270,6 +274,16 @@ def sagemaker_session_complete():
     return ims
 
 
+@pytest.fixture()
+def sagemaker_session_stopped():
+    boto_mock = Mock(name='boto_session')
+    boto_mock.client('logs').describe_log_streams.return_value = DEFAULT_LOG_STREAMS
+    boto_mock.client('logs').get_log_events.side_effect = DEFAULT_LOG_EVENTS
+    ims = sagemaker.Session(boto_session=boto_mock, sagemaker_client=Mock())
+    ims.sagemaker_client.describe_training_job.return_value = STOPPED_DESCRIBE_JOB_RESULT
+    return ims
+
+
 @pytest.fixture()
 def sagemaker_session_ready_lifecycle():
     boto_mock = Mock(name='boto_session')
@@ -302,6 +316,14 @@ def test_logs_for_job_no_wait(cw, sagemaker_session_complete):
     cw().assert_called_with(0, 'hi there #1')
 
 
+@patch('sagemaker.logs.ColorWrap')
+def test_logs_for_job_no_wait_stopped_job(cw, sagemaker_session_stopped):
+    ims = sagemaker_session_stopped
+    ims.logs_for_job(JOB_NAME)
+    ims.sagemaker_client.describe_training_job.assert_called_once_with(TrainingJobName=JOB_NAME)
+    cw().assert_called_with(0, 'hi there #1')
+
+
 @patch('sagemaker.logs.ColorWrap')
 def test_logs_for_job_wait_on_completed(cw, sagemaker_session_complete):
     ims = sagemaker_session_complete
@@ -310,6 +332,14 @@ def test_logs_for_job_wait_on_completed(cw, sagemaker_session_complete):
     cw().assert_called_with(0, 'hi there #1')
 
 
+@patch('sagemaker.logs.ColorWrap')
+def test_logs_for_job_wait_on_stopped(cw, sagemaker_session_stopped):
+    ims = sagemaker_session_stopped
+    ims.logs_for_job(JOB_NAME, wait=True, poll=0)
+    assert ims.sagemaker_client.describe_training_job.call_args_list == [call(TrainingJobName=JOB_NAME,)]
+    cw().assert_called_with(0, 'hi there #1')
+
+
 @patch('sagemaker.logs.ColorWrap')
 def test_logs_for_job_no_wait_on_running(cw, sagemaker_session_ready_lifecycle):
     ims = sagemaker_session_ready_lifecycle