def test_dispatcher_worker_retry(session_toy):
config = read_config(database_config_template())
event_config = read_config(ramp_config_template())
dispatcher = Dispatcher(config=config,
event_config=event_config,
worker=CondaEnvWorker,
n_workers=10,
hunger_policy='exit')
dispatcher.fetch_from_db(session_toy)
dispatcher.launch_workers(session_toy)
# Get one worker and set status to 'retry'
worker, (submission_id, submission_name) = \
dispatcher._processing_worker_queue.get()
setattr(worker, 'status', 'retry')
assert worker.status == 'retry'
# Add back to queue
dispatcher._processing_worker_queue.put_nowait(
(worker, (submission_id, submission_name)))
while not dispatcher._processing_worker_queue.empty():
dispatcher.collect_result(session_toy)
submissions = get_submissions(session_toy, 'iris_test', 'new')
assert submission_name in [sub[1] for sub in submissions]
def test_unit_test_dispatcher(session_toy):
# make sure that the size of the list is bigger than the number of
# submissions
config = read_config(database_config_template())
event_config = read_config(ramp_config_template())
dispatcher = Dispatcher(config=config,
event_config=event_config,
worker=CondaEnvWorker,
n_workers=100,
hunger_policy='exit')
# check that all the queue are empty
assert dispatcher._awaiting_worker_queue.empty()
assert dispatcher._processing_worker_queue.empty()
assert dispatcher._processed_submission_queue.empty()
# check that all submissions are queued
submissions = get_submissions(session_toy, 'iris_test', 'new')
dispatcher.fetch_from_db(session_toy)
# we should remove the starting kit from the length of the submissions for
# each user
assert dispatcher._awaiting_worker_queue.qsize() == len(submissions) - 2
submissions = get_submissions(session_toy, 'iris_test', 'sent_to_training')
assert len(submissions) == 6
# start the training
dispatcher.launch_workers(session_toy)
# be sure that the training is finished
while not dispatcher._processing_worker_queue.empty():
dispatcher.collect_result(session_toy)
assert len(get_submissions(session_toy, 'iris_test', 'new')) == 2
assert (len(get_submissions(session_toy, 'iris_test',
'training_error')) == 2)
assert len(get_submissions(session_toy, 'iris_test', 'tested')) == 4
dispatcher.update_database_results(session_toy)
assert dispatcher._processed_submission_queue.empty()
event = get_event(session_toy, 'iris_test')
assert event.private_leaderboard_html
assert event.public_leaderboard_html_with_links
assert event.public_leaderboard_html_no_links
assert event.failed_leaderboard_html
assert event.new_leaderboard_html is None
assert event.public_competition_leaderboard_html
assert event.private_competition_leaderboard_html
def test_dispatcher_aws_not_launching(session_toy_aws, caplog):
# given the test config file the instance should not be able to launch
# due to authentication error
# after unsuccessful try the worker should teardown
config = read_config(database_config_template())
event_config = read_config(ramp_aws_config_template())
dispatcher = Dispatcher(config=config,
event_config=event_config,
worker=AWSWorker,
n_workers=10,
hunger_policy='exit')
dispatcher.fetch_from_db(session_toy_aws)
submissions = get_submissions(session_toy_aws, 'iris_aws_test', 'new')
dispatcher.launch_workers(session_toy_aws)
assert 'AuthFailure' in caplog.text
# training should not have started
assert 'training' not in caplog.text
num_running_workers = dispatcher._processing_worker_queue.qsize()
assert num_running_workers == 0
submissions2 = get_submissions(session_toy_aws, 'iris_aws_test', 'new')
# assert that all the submissions are still in the 'new' state
assert len(submissions) == len(submissions2)
def test_info_on_training_error(test_launch_ec2_instances, upload_submission,
launch_train, is_spot_terminated,
training_finished, training_successful,
get_log_content, check_instance_status,
download_log, session_toy_aws, caplog):
# make sure that the Python error from the solution is passed to the
# dispatcher
# everything shoud be mocked as correct output from AWS instances
# on setting up the instance and loading the submission
# mock dummy AWS instance
class DummyInstance:
id = 1
test_launch_ec2_instances.return_value = (DummyInstance(), ), 0
upload_submission.return_value = 0
launch_train.return_value = 0
is_spot_terminated.return_value = 0
training_finished.return_value = False
download_log.return_value = 0
config = read_config(database_config_template())
event_config = read_config(ramp_aws_config_template())
dispatcher = Dispatcher(config=config,
event_config=event_config,
worker=AWSWorker,
n_workers=10,
hunger_policy='exit')
dispatcher.fetch_from_db(session_toy_aws)
dispatcher.launch_workers(session_toy_aws)
num_running_workers = dispatcher._processing_worker_queue.qsize()
# worker, (submission_id, submission_name) = \
# dispatcher._processing_worker_queue.get()
# assert worker.status == 'running'
submissions = get_submissions(session_toy_aws, 'iris_aws_test', 'training')
ids = [submissions[idx][0] for idx in range(len(submissions))]
assert len(submissions) > 1
assert num_running_workers == len(ids)
dispatcher.time_between_collection = 0
training_successful.return_value = False
# now we will end the submission with training error
training_finished.return_value = True
training_error_msg = 'Python error here'
get_log_content.return_value = training_error_msg
check_instance_status.return_value = 'finished'
dispatcher.collect_result(session_toy_aws)
# the worker which we were using should have been teared down
num_running_workers = dispatcher._processing_worker_queue.qsize()
assert num_running_workers == 0
submissions = get_submissions(session_toy_aws, 'iris_aws_test',
'training_error')
assert len(submissions) == len(ids)
submission = get_submission_by_id(session_toy_aws, submissions[0][0])
assert training_error_msg in submission.error_msg