def task(work_dir, state_path, random_file_path):
"""
Runs the end-to-end workflow of the Dependency Manager.
Note: ProcessPoolExecutor must serialize everything before sending it to the worker,
so this function needs to be defined at the top-level.
# """
# Mock Bundle Service to return a random file object
mock_bundle_service = MagicMock()
mock_bundle_service.get_bundle_info = MagicMock(
return_value={'type': "file"})
file_obj = open(random_file_path, "rb")
mock_bundle_service.get_bundle_contents = MagicMock(return_value=file_obj)
# Create and start a dependency manager
process_id = os.getpid()
print(f"{process_id}: Starting a DependencyManager...")
dependency_manager = DependencyManager(
commit_file=state_path,
bundle_service=mock_bundle_service,
worker_dir=work_dir,
max_cache_size_bytes=2048,
download_dependencies_max_retries=1,
)
dependency_manager.start()
print(f"{process_id}: Started with work directory: {work_dir}.")
# Register a run's UUID as a dependent of a parent bundle with UUID 0x1
dependency_key = DependencyKey(parent_uuid="0x1", parent_path="parent")
run_uuid = f"0x{process_id}"
state = dependency_manager.get(run_uuid, dependency_key)
assert (run_uuid in state.dependents
), f"{process_id}: Expected {run_uuid} as one of the dependents."
# Release the run bundle as a dependent
dependency_manager.release(run_uuid, dependency_key)
dependencies = dependency_manager._fetch_dependencies()
if dependency_key in dependencies:
state = dependencies[dependency_key]
print(f"{process_id}: Checking {run_uuid} in {state.dependents}")
assert (
run_uuid not in state.dependents
), f"{process_id}: Dependent should not be in the list of dependents after unregistering."
# Keep the dependency manager running for some time to test the loop
time.sleep(30)
# Stop the Dependency Manager
print(f"{process_id}: Stopping DependencyManger...")
dependency_manager.stop()
print(f"{process_id}: Done.")
def setUp(self):
if module_failed:
self.skipTest('Issue with ratarmountcore.')
self.work_dir = tempfile.mkdtemp()
self.state_path = os.path.join(self.work_dir,
"dependencies-state.json")
self.dependency_manager = DependencyManager(
commit_file=self.state_path,
bundle_service=None,
worker_dir=self.work_dir,
max_cache_size_bytes=1024,
download_dependencies_max_retries=1,
)
def main():
args = parse_args()
# This quits if connection unsuccessful
bundle_service = connect_to_codalab_server(args.server, args.password_file)
# Configure logging
logging.basicConfig(
format='%(asctime)s %(message)s',
level=(logging.DEBUG if args.verbose else logging.INFO))
if args.shared_file_system:
# No need to store bundles locally if filesystems are shared
local_bundles_dir = None
# Also no need to download dependencies if they're on the filesystem already
dependency_manager = None
else:
local_bundles_dir = os.path.join(args.work_dir, 'runs')
dependency_manager = DependencyManager(
os.path.join(args.work_dir, 'dependencies-state.json'),
bundle_service,
args.work_dir,
args.max_work_dir_size,
)
# Set up local directories
if not os.path.exists(args.work_dir):
logging.debug('Work dir %s doesn\'t exist, creating.', args.work_dir)
os.makedirs(args.work_dir, 0o770)
if local_bundles_dir and not os.path.exists(local_bundles_dir):
logger.info('%s doesn\'t exist, creating.', local_bundles_dir)
os.makedirs(local_bundles_dir, 0o770)
docker_runtime = docker_utils.get_available_runtime()
image_manager = DockerImageManager(
os.path.join(args.work_dir, 'images-state.json'),
args.max_image_cache_size,
args.max_image_size,
)
worker = Worker(
image_manager,
dependency_manager,
os.path.join(args.work_dir, 'worker-state.json'),
args.cpuset,
args.gpuset,
args.id,
args.tag,
args.work_dir,
local_bundles_dir,
args.exit_when_idle,
args.idle_seconds,
bundle_service,
args.shared_file_system,
docker_runtime=docker_runtime,
docker_network_prefix=args.network_prefix,
)
# Register a signal handler to ensure safe shutdown.
for sig in [signal.SIGTERM, signal.SIGINT, signal.SIGHUP]:
signal.signal(sig, lambda signup, frame: worker.signal())
# BEGIN: DO NOT CHANGE THIS LINE UNLESS YOU KNOW WHAT YOU ARE DOING
# THIS IS HERE TO KEEP TEST-CLI FROM HANGING
logger.info('Worker started!')
# END
worker.start()
def main():
args = parse_args()
if args.tag and not args.tag.replace("-", "").isalnum():
raise argparse.ArgumentTypeError(
"Worker tag must only contain letters, numbers or hyphens."
)
# Configure logging
log_format: str = '%(asctime)s %(message)s'
if args.verbose:
log_format += ' %(pathname)s %(lineno)d'
log_level = logging.DEBUG
else:
log_level = logging.INFO
logging.basicConfig(format=log_format, level=log_level)
logging.getLogger('urllib3').setLevel(logging.INFO)
# Initialize sentry logging
if using_sentry():
initialize_sentry()
# This quits if connection unsuccessful
bundle_service = connect_to_codalab_server(args.server, args.password_file)
# Load some data into sentry
if using_sentry():
load_sentry_data(username=bundle_service._username, **vars(args))
if args.shared_file_system:
# No need to store bundles locally if filesystems are shared
local_bundles_dir = None
# Also no need to download dependencies if they're on the filesystem already
dependency_manager = None
else:
local_bundles_dir = os.path.join(args.work_dir, 'runs')
dependency_manager = DependencyManager(
os.path.join(args.work_dir, 'dependencies-state.json'),
bundle_service,
args.work_dir,
args.max_work_dir_size,
args.download_dependencies_max_retries,
)
if args.container_runtime == "singularity":
singularity_folder = os.path.join(args.work_dir, 'codalab_singularity_images')
if not os.path.exists(singularity_folder):
logger.info(
'Local singularity image location %s doesn\'t exist, creating.', singularity_folder,
)
os.makedirs(singularity_folder, 0o770)
image_manager = SingularityImageManager(
args.max_image_size, args.max_image_cache_size, singularity_folder,
)
# todo workers with singularity don't work because this is set to none -- handle this
docker_runtime = None
else:
image_manager = DockerImageManager(
os.path.join(args.work_dir, 'images-state.json'),
args.max_image_cache_size,
args.max_image_size,
)
docker_runtime = docker_utils.get_available_runtime()
# Set up local directories
if not os.path.exists(args.work_dir):
logging.debug('Work dir %s doesn\'t exist, creating.', args.work_dir)
os.makedirs(args.work_dir, 0o770)
if local_bundles_dir and not os.path.exists(local_bundles_dir):
logger.info('%s doesn\'t exist, creating.', local_bundles_dir)
os.makedirs(local_bundles_dir, 0o770)
worker = Worker(
image_manager,
dependency_manager,
# Include the worker ID in the worker state JSON path, so multiple workers
# sharing the same work directory maintain their own state.
os.path.join(args.work_dir, f'worker-state-{args.id}.json'),
args.cpuset,
args.gpuset,
args.max_memory,
args.id,
args.tag,
args.work_dir,
local_bundles_dir,
args.exit_when_idle,
args.exit_after_num_runs,
args.idle_seconds,
args.checkin_frequency_seconds,
bundle_service,
args.shared_file_system,
args.tag_exclusive,
args.group,
docker_runtime=docker_runtime,
docker_network_prefix=args.network_prefix,
pass_down_termination=args.pass_down_termination,
delete_work_dir_on_exit=args.delete_work_dir_on_exit,
exit_on_exception=args.exit_on_exception,
shared_memory_size_gb=args.shared_memory_size_gb,
preemptible=args.preemptible,
)
# Register a signal handler to ensure safe shutdown.
for sig in [signal.SIGTERM, signal.SIGINT, signal.SIGHUP]:
signal.signal(sig, lambda signup, frame: worker.signal())
# BEGIN: DO NOT CHANGE THIS LINE UNLESS YOU KNOW WHAT YOU ARE DOING
# THIS IS HERE TO KEEP TEST-CLI FROM HANGING
logger.info('Worker started!')
# END
worker.start()
class DependencyManagerTest(unittest.TestCase):
def setUp(self):
if module_failed:
self.skipTest('Issue with ratarmountcore.')
self.work_dir = tempfile.mkdtemp()
self.state_path = os.path.join(self.work_dir,
"dependencies-state.json")
self.dependency_manager = DependencyManager(
commit_file=self.state_path,
bundle_service=None,
worker_dir=self.work_dir,
max_cache_size_bytes=1024,
download_dependencies_max_retries=1,
)
def tearDown(self):
shutil.rmtree(self.work_dir)
def test_get_has(self):
dependent_uuid = "0x2"
dependency_key = DependencyKey(parent_uuid="0x1", parent_path="parent")
state = self.dependency_manager.get(dependent_uuid, dependency_key)
self.assertTrue(self.dependency_manager.has(dependency_key))
self.assertEqual(state.stage, "DOWNLOADING")
self.assertEqual(state.path, "0x1_parent")
self.assertEqual(state.dependents, {dependent_uuid})
def test_release(self):
dependency_key = DependencyKey(parent_uuid="0x1", parent_path="parent")
self.dependency_manager.get("0x2", dependency_key)
state = self.dependency_manager.get("0x3", dependency_key)
# Passing in the same dependency key with a different dependent, will just add the dependent
self.assertEqual(state.dependents, {"0x2", "0x3"})
# Release 0x2 as a dependent
self.dependency_manager.release("0x2", dependency_key)
with self.dependency_manager._state_lock:
dependencies = self.dependency_manager._fetch_dependencies()
state = dependencies[dependency_key]
self.assertEqual(state.dependents, {"0x3"})
# Release 0x3 as a dependent - should be left with no dependents
self.dependency_manager.release("0x3", dependency_key)
with self.dependency_manager._state_lock:
dependencies = self.dependency_manager._fetch_dependencies()
state = dependencies[dependency_key]
self.assertEqual(len(state.dependents), 0)
def test_all_dependencies(self):
dependency_key = DependencyKey(parent_uuid="0x1", parent_path="parent")
self.dependency_manager.get("0x2", dependency_key)
dependency_key = DependencyKey(parent_uuid="0x3",
parent_path="parent2")
self.dependency_manager.get("0x4", dependency_key)
dependency_keys = self.dependency_manager.all_dependencies
self.assertEqual(len(dependency_keys), 2)
@unittest.skip("Flufl.lock doesn't seem to work on GHA for some reason, "
"even though this test passes on other machines.")
def test_concurrency(self):
num_of_dependency_managers = 10
executor = ProcessPoolExecutor(max_workers=num_of_dependency_managers)
random_file_path = os.path.join(self.work_dir, "random_file")
with open(random_file_path, "wb") as f:
f.seek((1024 * 1024 * 1024) - 1) # 1 GB
f.write(b"\0")
futures = [
executor.submit(task, self.work_dir, self.state_path,
random_file_path)
for _ in range(num_of_dependency_managers)
]
for future in futures:
print(future.result())
self.assertIsNone(future.exception())
executor.shutdown()
def main():
args = parse_args()
if args.tag and not args.tag.isalnum():
raise argparse.ArgumentTypeError(
"Worker tag must be alphanumeric (only contain letters and numbers)."
)
# Configure logging
logging.basicConfig(
format='%(asctime)s %(message)s %(pathname)s %(lineno)d',
level=(logging.DEBUG if args.verbose else logging.INFO),
)
logging.getLogger('urllib3').setLevel(logging.INFO)
# Initialize sentry logging
if using_sentry():
initialize_sentry()
# This quits if connection unsuccessful
bundle_service = connect_to_codalab_server(args.server, args.password_file)
# Load some data into sentry
if using_sentry():
load_sentry_data(username=bundle_service._username, **vars(args))
if args.shared_file_system:
# No need to store bundles locally if filesystems are shared
local_bundles_dir = None
# Also no need to download dependencies if they're on the filesystem already
dependency_manager = None
else:
local_bundles_dir = os.path.join(args.work_dir, 'runs')
dependency_manager = DependencyManager(
os.path.join(args.work_dir, 'dependencies-state.json'),
bundle_service,
args.work_dir,
args.max_work_dir_size,
)
# Set up local directories
if not os.path.exists(args.work_dir):
logging.debug('Work dir %s doesn\'t exist, creating.', args.work_dir)
os.makedirs(args.work_dir, 0o770)
if local_bundles_dir and not os.path.exists(local_bundles_dir):
logger.info('%s doesn\'t exist, creating.', local_bundles_dir)
os.makedirs(local_bundles_dir, 0o770)
docker_runtime = docker_utils.get_available_runtime()
image_manager = DockerImageManager(
os.path.join(args.work_dir, 'images-state.json'),
args.max_image_cache_size,
args.max_image_size,
)
worker = Worker(
image_manager,
dependency_manager,
os.path.join(args.work_dir, 'worker-state.json'),
args.cpuset,
args.gpuset,
args.max_memory,
args.id,
args.tag,
args.work_dir,
local_bundles_dir,
args.exit_when_idle,
args.exit_after_num_runs,
args.idle_seconds,
args.checkin_frequency_seconds,
bundle_service,
args.shared_file_system,
args.tag_exclusive,
args.group,
docker_runtime=docker_runtime,
docker_network_prefix=args.network_prefix,
pass_down_termination=args.pass_down_termination,
delete_work_dir_on_exit=args.delete_work_dir_on_exit,
exit_on_exception=args.exit_on_exception,
)
# Register a signal handler to ensure safe shutdown.
for sig in [signal.SIGTERM, signal.SIGINT, signal.SIGHUP]:
signal.signal(sig, lambda signup, frame: worker.signal())
# BEGIN: DO NOT CHANGE THIS LINE UNLESS YOU KNOW WHAT YOU ARE DOING
# THIS IS HERE TO KEEP TEST-CLI FROM HANGING
logger.info('Worker started!')
# END
worker.start()