def setUp(self):
self.data_dir = tempfile.mkdtemp()
database_path = os.path.join(self.data_dir, "mephisto.db")
self.db = LocalMephistoDB(database_path)
self.task_id = self.db.new_task("test_mock",
MockBlueprint.BLUEPRINT_TYPE)
self.task_run_id = get_test_task_run(self.db)
self.task_run = TaskRun(self.db, self.task_run_id)
architect_config = OmegaConf.structured(
MephistoConfig(
architect=MockArchitectArgs(should_run_server=True), ))
self.architect = MockArchitect(self.db, architect_config, EMPTY_STATE,
self.task_run, self.data_dir)
self.architect.prepare()
self.architect.deploy()
self.urls = self.architect._get_socket_urls() # FIXME
self.url = self.urls[0]
self.provider = MockProvider(self.db)
self.provider.setup_resources_for_task_run(self.task_run,
self.task_run.args,
self.url)
self.launcher = TaskLauncher(self.db, self.task_run,
self.get_mock_assignment_data_array())
self.launcher.create_assignments()
self.launcher.launch_units(self.url)
self.sup = None
def test_launch_assignments_with_concurrent_unit_cap(self):
"""Initialize a launcher on a task run, then create the assignments"""
cap_values = [1, 2, 3, 4, 5]
for max_num_units in cap_values:
mock_data_array = self.get_mock_assignment_data_array()
launcher = TaskLauncher(
self.db,
self.task_run,
mock_data_array,
max_num_concurrent_units=max_num_units,
)
launcher.launched_units = LimitedDict(
launcher.max_num_concurrent_units)
launcher.create_assignments()
launcher.launch_units("dummy-url:3000")
start_time = time.time()
while set([u.get_status() for u in launcher.units
]) != {AssignmentState.COMPLETED}:
for unit in launcher.units:
if unit.get_status() == AssignmentState.LAUNCHED:
unit.set_db_status(AssignmentState.COMPLETED)
time.sleep(0.1)
self.assertEqual(launcher.launched_units.exceed_limit, False)
curr_time = time.time()
self.assertLessEqual(curr_time - start_time,
MAX_WAIT_TIME_UNIT_LAUNCH)
launcher.expire_units()
self.tearDown()
self.setUp()
def test_assignments_generator(self):
"""Initialize a launcher on a task run, then try generate the assignments"""
mock_data_array = self.get_mock_assignment_data_generator()
start_time = time.time()
launcher = TaskLauncher(self.db, self.task_run, mock_data_array)
launcher.create_assignments()
end_time = time.time()
self.assertLessEqual(
end_time - start_time,
(NUM_GENERATED_ASSIGNMENTS * WAIT_TIME_TILL_NEXT_ASSIGNMENT) / 2,
)
def test_init_on_task_run(self):
"""Initialize a launcher on a task_run"""
launcher = TaskLauncher(self.db, self.task_run,
self.get_mock_assignment_data_array())
self.assertEqual(self.db, launcher.db)
self.assertEqual(self.task_run, launcher.task_run)
self.assertEqual(len(launcher.assignments), 0)
self.assertEqual(len(launcher.units), 0)
self.assertEqual(launcher.provider_type, MockProvider.PROVIDER_TYPE)
def test_create_launch_expire_assignments(self):
"""Initialize a launcher on a task run, then create the assignments"""
mock_data_array = self.get_mock_assignment_data_array()
launcher = TaskLauncher(self.db, self.task_run, mock_data_array)
launcher.create_assignments()
self.assertEqual(
len(launcher.assignments),
len(mock_data_array),
"Inequal number of assignments existed than were launched",
)
self.assertEqual(
len(launcher.units),
len(mock_data_array) * len(mock_data_array[0]["unit_data"]),
"Inequal number of units created than were expected",
)
for unit in launcher.units:
self.assertEqual(unit.get_db_status(), AssignmentState.CREATED)
for assignment in launcher.assignments:
self.assertEqual(assignment.get_status(), AssignmentState.CREATED)
launcher.launch_units("dummy-url:3000")
for unit in launcher.units:
self.assertEqual(unit.get_db_status(), AssignmentState.LAUNCHED)
time.sleep(WAIT_TIME_TILL_NEXT_UNIT)
for assignment in launcher.assignments:
self.assertEqual(assignment.get_status(), AssignmentState.LAUNCHED)
launcher.expire_units()
for unit in launcher.units:
self.assertEqual(unit.get_db_status(), AssignmentState.EXPIRED)
for assignment in launcher.assignments:
self.assertEqual(assignment.get_status(), AssignmentState.EXPIRED)
def parse_and_launch_run(
self,
arg_list: Optional[List[str]] = None,
extra_args: Optional[Dict[str, Any]] = None,
) -> str:
"""
Parse the given arguments and launch a job.
"""
if extra_args is None:
extra_args = {}
# Extract the abstractions being used
parser = self._get_baseline_argparser()
type_args, task_args_string = parser.parse_known_args(arg_list)
requesters = self.db.find_requesters(requester_name=type_args.requester_name)
if len(requesters) == 0:
raise EntryDoesNotExistException(
f"No requester found with name {type_args.requester_name}"
)
requester = requesters[0]
requester_id = requester.db_id
provider_type = requester.provider_type
# Parse the arguments for the abstractions to ensure
# everything required is set
BlueprintClass = get_blueprint_from_type(type_args.blueprint_type)
ArchitectClass = get_architect_from_type(type_args.architect_type)
CrowdProviderClass = get_crowd_provider_from_type(provider_type)
task_args, _unknown = self._parse_args_from_classes(
BlueprintClass, ArchitectClass, CrowdProviderClass, task_args_string
)
task_args.update(extra_args)
# Load the classes to force argument validation before anything
# is actually created in the database
# TODO(#94) perhaps parse the arguments for these things one at a time?
BlueprintClass.assert_task_args(task_args)
ArchitectClass.assert_task_args(task_args)
CrowdProviderClass.assert_task_args(task_args)
# Find an existing task or create a new one
task_name = task_args.get("task_name")
if task_name is None:
task_name = type_args.blueprint_type
logger.warning(
f"Task is using the default blueprint name {task_name} as a name, as no task_name is provided"
)
tasks = self.db.find_tasks(task_name=task_name)
task_id = None
if len(tasks) == 0:
task_id = self.db.new_task(task_name, type_args.blueprint_type)
else:
task_id = tasks[0].db_id
logger.info(f"Creating a task run under task name: {task_name}")
# Create a new task run
new_run_id = self.db.new_task_run(
task_id,
requester_id,
" ".join([shlex.quote(x) for x in task_args_string]),
provider_type,
type_args.blueprint_type,
requester.is_sandbox(),
)
task_run = TaskRun(self.db, new_run_id)
try:
# If anything fails after here, we have to cleanup the architect
build_dir = os.path.join(task_run.get_run_dir(), "build")
os.makedirs(build_dir, exist_ok=True)
architect = ArchitectClass(self.db, task_args, task_run, build_dir)
# Register the blueprint with args to the task run,
# ensure cached
blueprint = BlueprintClass(task_run, task_args)
task_run.get_blueprint(opts=task_args)
# Setup and deploy the server
built_dir = architect.prepare()
task_url = architect.deploy()
# TODO(#102) maybe the cleanup (destruction of the server configuration?) should only
# happen after everything has already been reviewed, this way it's possible to
# retrieve the exact build directory to review a task for real
architect.cleanup()
# Create the backend runner
task_runner = BlueprintClass.TaskRunnerClass(task_run, task_args)
# Small hack for auto appending block qualification
existing_qualifications = task_args.get("qualifications", [])
if task_args.get("block_qualification") is not None:
existing_qualifications.append(
make_qualification_dict(
task_args["block_qualification"], QUAL_NOT_EXIST, None
)
)
if task_args.get("onboarding_qualification") is not None:
existing_qualifications.append(
make_qualification_dict(
OnboardingRequired.get_failed_qual(
task_args["onboarding_qualification"]
),
QUAL_NOT_EXIST,
None,
)
)
task_args["qualifications"] = existing_qualifications
# Register the task with the provider
provider = CrowdProviderClass(self.db)
provider.setup_resources_for_task_run(task_run, task_args, task_url)
initialization_data_array = blueprint.get_initialization_data()
# Link the job together
job = self.supervisor.register_job(
architect, task_runner, provider, existing_qualifications
)
if self.supervisor.sending_thread is None:
self.supervisor.launch_sending_thread()
except (KeyboardInterrupt, Exception) as e:
logger.error(
"Encountered error while launching run, shutting down", exc_info=True
)
try:
architect.shutdown()
except (KeyboardInterrupt, Exception) as architect_exception:
logger.exception(
f"Could not shut down architect: {architect_exception}",
exc_info=True,
)
raise e
launcher = TaskLauncher(self.db, task_run, initialization_data_array)
launcher.create_assignments()
launcher.launch_units(task_url)
self._task_runs_tracked[task_run.db_id] = TrackedRun(
task_run=task_run,
task_launcher=launcher,
task_runner=task_runner,
architect=architect,
job=job,
)
return task_run.db_id
def validate_and_run_config_or_die(
self,
run_config: DictConfig,
shared_state: Optional[SharedTaskState] = None,
) -> str:
"""
Parse the given arguments and launch a job.
"""
if shared_state is None:
shared_state = SharedTaskState()
# First try to find the requester:
requester_name = run_config.provider.requester_name
requesters = self.db.find_requesters(requester_name=requester_name)
if len(requesters) == 0:
if run_config.provider.requester_name == "MOCK_REQUESTER":
requesters = [get_mock_requester(self.db)]
else:
raise EntryDoesNotExistException(
f"No requester found with name {requester_name}")
requester = requesters[0]
requester_id = requester.db_id
provider_type = requester.provider_type
assert provider_type == run_config.provider._provider_type, (
f"Found requester for name {requester_name} is not "
f"of the specified type {run_config.provider._provider_type}, "
f"but is instead {provider_type}.")
# Next get the abstraction classes, and run validation
# before anything is actually created in the database
blueprint_type = run_config.blueprint._blueprint_type
architect_type = run_config.architect._architect_type
BlueprintClass = get_blueprint_from_type(blueprint_type)
ArchitectClass = get_architect_from_type(architect_type)
CrowdProviderClass = get_crowd_provider_from_type(provider_type)
BlueprintClass.assert_task_args(run_config, shared_state)
ArchitectClass.assert_task_args(run_config, shared_state)
CrowdProviderClass.assert_task_args(run_config, shared_state)
# Find an existing task or create a new one
task_name = run_config.task.get("task_name", None)
if task_name is None:
task_name = blueprint_type
logger.warning(
f"Task is using the default blueprint name {task_name} as a name, "
"as no task_name is provided")
tasks = self.db.find_tasks(task_name=task_name)
task_id = None
if len(tasks) == 0:
task_id = self.db.new_task(task_name, blueprint_type)
else:
task_id = tasks[0].db_id
logger.info(f"Creating a task run under task name: {task_name}")
# Create a new task run
new_run_id = self.db.new_task_run(
task_id,
requester_id,
json.dumps(OmegaConf.to_container(run_config, resolve=True)),
provider_type,
blueprint_type,
requester.is_sandbox(),
)
task_run = TaskRun(self.db, new_run_id)
try:
# If anything fails after here, we have to cleanup the architect
build_dir = os.path.join(task_run.get_run_dir(), "build")
os.makedirs(build_dir, exist_ok=True)
architect = ArchitectClass(self.db, run_config, shared_state,
task_run, build_dir)
# Register the blueprint with args to the task run,
# ensure cached
blueprint = BlueprintClass(task_run, run_config, shared_state)
task_run.get_blueprint(args=run_config, shared_state=shared_state)
# Setup and deploy the server
built_dir = architect.prepare()
task_url = architect.deploy()
# TODO(#102) maybe the cleanup (destruction of the server configuration?) should only
# happen after everything has already been reviewed, this way it's possible to
# retrieve the exact build directory to review a task for real
architect.cleanup()
# Create the backend runner
task_runner = BlueprintClass.TaskRunnerClass(
task_run, run_config, shared_state)
# Small hack for auto appending block qualification
existing_qualifications = shared_state.qualifications
if run_config.blueprint.get("block_qualification",
None) is not None:
existing_qualifications.append(
make_qualification_dict(
run_config.blueprint.block_qualification,
QUAL_NOT_EXIST, None))
if run_config.blueprint.get("onboarding_qualification",
None) is not None:
existing_qualifications.append(
make_qualification_dict(
OnboardingRequired.get_failed_qual(
run_config.blueprint.onboarding_qualification, ),
QUAL_NOT_EXIST,
None,
))
shared_state.qualifications = existing_qualifications
# Register the task with the provider
provider = CrowdProviderClass(self.db)
provider.setup_resources_for_task_run(task_run, run_config,
task_url)
initialization_data_array = blueprint.get_initialization_data()
# Link the job together
job = self.supervisor.register_job(architect, task_runner,
provider,
existing_qualifications)
if self.supervisor.sending_thread is None:
self.supervisor.launch_sending_thread()
except (KeyboardInterrupt, Exception) as e:
logger.error(
"Encountered error while launching run, shutting down",
exc_info=True)
try:
architect.shutdown()
except (KeyboardInterrupt, Exception) as architect_exception:
logger.exception(
f"Could not shut down architect: {architect_exception}",
exc_info=True,
)
raise e
launcher = TaskLauncher(self.db, task_run, initialization_data_array)
launcher.create_assignments()
launcher.launch_units(task_url)
self._task_runs_tracked[task_run.db_id] = TrackedRun(
task_run=task_run,
task_launcher=launcher,
task_runner=task_runner,
architect=architect,
job=job,
)
task_run.update_completion_progress(status=False)
return task_run.db_id
class TestSupervisor(unittest.TestCase):
"""
Unit testing for the Mephisto Supervisor,
uses WebsocketChannel and MockArchitect
"""
def setUp(self):
self.data_dir = tempfile.mkdtemp()
database_path = os.path.join(self.data_dir, "mephisto.db")
self.db = LocalMephistoDB(database_path)
self.task_id = self.db.new_task("test_mock",
MockBlueprint.BLUEPRINT_TYPE)
self.task_run_id = get_test_task_run(self.db)
self.task_run = TaskRun(self.db, self.task_run_id)
architect_config = OmegaConf.structured(
MephistoConfig(
architect=MockArchitectArgs(should_run_server=True), ))
self.architect = MockArchitect(self.db, architect_config, EMPTY_STATE,
self.task_run, self.data_dir)
self.architect.prepare()
self.architect.deploy()
self.urls = self.architect._get_socket_urls() # FIXME
self.url = self.urls[0]
self.provider = MockProvider(self.db)
self.provider.setup_resources_for_task_run(self.task_run,
self.task_run.args,
self.url)
self.launcher = TaskLauncher(self.db, self.task_run,
self.get_mock_assignment_data_array())
self.launcher.create_assignments()
self.launcher.launch_units(self.url)
self.sup = None
def tearDown(self):
if self.sup is not None:
self.sup.shutdown()
self.launcher.expire_units()
self.architect.cleanup()
self.architect.shutdown()
self.db.shutdown()
shutil.rmtree(self.data_dir)
def get_mock_assignment_data_array(self) -> List[InitializationData]:
mock_data = MockTaskRunner.get_mock_assignment_data()
return [mock_data, mock_data]
def test_initialize_supervisor(self):
"""Ensure that the supervisor object can even be created"""
sup = Supervisor(self.db)
self.assertIsNotNone(sup)
self.assertDictEqual(sup.agents, {})
self.assertDictEqual(sup.channels, {})
sup.shutdown()
def test_channel_operations(self):
"""
Initialize a channel, and ensure the basic
startup and shutdown functions are working
"""
sup = Supervisor(self.db)
self.sup = sup
TaskRunnerClass = MockBlueprint.TaskRunnerClass
args = MockBlueprint.ArgsClass()
config = OmegaConf.structured(MephistoConfig(blueprint=args))
task_runner = TaskRunnerClass(self.task_run, config, EMPTY_STATE)
test_job = Job(
architect=self.architect,
task_runner=task_runner,
provider=self.provider,
qualifications=[],
registered_channel_ids=[],
)
channels = self.architect.get_channels(sup._on_channel_open,
sup._on_catastrophic_disconnect,
sup._on_message)
channel = channels[0]
channel.open()
channel_id = channel.channel_id
self.assertIsNotNone(channel_id)
channel.close()
self.assertTrue(channel.is_closed())
def test_register_concurrent_job(self):
"""Test registering and running a job that requires multiple workers"""
# Handle baseline setup
sup = Supervisor(self.db)
self.sup = sup
TaskRunnerClass = MockBlueprint.TaskRunnerClass
args = MockBlueprint.ArgsClass()
args.timeout_time = 5
args.is_concurrent = False
config = OmegaConf.structured(MephistoConfig(blueprint=args))
task_runner = TaskRunnerClass(self.task_run, config, EMPTY_STATE)
sup.register_job(self.architect, task_runner, self.provider)
self.assertEqual(len(sup.channels), 1)
channel_info = list(sup.channels.values())[0]
self.assertIsNotNone(channel_info)
self.assertTrue(channel_info.channel.is_alive)
channel_id = channel_info.channel_id
task_runner = channel_info.job.task_runner
self.assertIsNotNone(channel_id)
self.assertEqual(
len(self.architect.server.subs),
1,
"MockServer doesn't see registered channel",
)
self.assertIsNotNone(
self.architect.server.last_alive_packet,
"No alive packet received by server",
)
sup.launch_sending_thread()
self.assertIsNotNone(sup.sending_thread)
# Register a worker
mock_worker_name = "MOCK_WORKER"
self.architect.server.register_mock_worker(mock_worker_name)
workers = self.db.find_workers(worker_name=mock_worker_name)
self.assertEqual(len(workers), 1, "Worker not successfully registered")
worker = workers[0]
self.architect.server.register_mock_worker(mock_worker_name)
workers = self.db.find_workers(worker_name=mock_worker_name)
self.assertEqual(len(workers), 1, "Worker potentially re-registered")
worker_id = workers[0].db_id
self.assertEqual(len(task_runner.running_assignments), 0)
# Register an agent
mock_agent_details = "FAKE_ASSIGNMENT"
self.architect.server.register_mock_agent(worker_id,
mock_agent_details)
agents = self.db.find_agents()
self.assertEqual(len(agents), 1, "Agent was not created properly")
self.architect.server.register_mock_agent(worker_id,
mock_agent_details)
agents = self.db.find_agents()
self.assertEqual(len(agents), 1, "Agent may have been duplicated")
agent = agents[0]
self.assertIsNotNone(agent)
self.assertEqual(len(sup.agents), 1,
"Agent not registered with supervisor")
self.assertEqual(len(task_runner.running_units), 1,
"Ready task was not launched")
# Register another worker
mock_worker_name = "MOCK_WORKER_2"
self.architect.server.register_mock_worker(mock_worker_name)
workers = self.db.find_workers(worker_name=mock_worker_name)
worker_id = workers[0].db_id
# Register an agent
mock_agent_details = "FAKE_ASSIGNMENT_2"
self.architect.server.register_mock_agent(worker_id,
mock_agent_details)
self.assertEqual(len(task_runner.running_units), 2,
"Tasks were not launched")
agents = [a.agent for a in sup.agents.values()]
# Make both agents act
agent_id_1, agent_id_2 = agents[0].db_id, agents[1].db_id
agent_1_data = agents[0].datastore.agent_data[agent_id_1]
agent_2_data = agents[1].datastore.agent_data[agent_id_2]
self.architect.server.send_agent_act(agent_id_1, {"text": "message1"})
self.architect.server.send_agent_act(agent_id_2, {"text": "message2"})
# Give up to 1 seconds for the actual operations to occur
start_time = time.time()
TIMEOUT_TIME = 1
while time.time() - start_time < TIMEOUT_TIME:
if len(agent_1_data["acts"]) > 0:
break
time.sleep(0.1)
self.assertLess(time.time() - start_time, TIMEOUT_TIME,
"Did not process messages in time")
# Give up to 1 seconds for the task to complete afterwards
start_time = time.time()
TIMEOUT_TIME = 1
while time.time() - start_time < TIMEOUT_TIME:
if len(task_runner.running_units) == 0:
break
time.sleep(0.1)
self.assertLess(time.time() - start_time, TIMEOUT_TIME,
"Did not complete task in time")
# Give up to 1 seconds for all messages to propogate
start_time = time.time()
TIMEOUT_TIME = 1
while time.time() - start_time < TIMEOUT_TIME:
if self.architect.server.actions_observed == 2:
break
time.sleep(0.1)
self.assertLess(time.time() - start_time, TIMEOUT_TIME,
"Not all actions observed in time")
sup.shutdown()
self.assertTrue(channel_info.channel.is_closed)
def test_register_job(self):
"""Test registering and running a job run asynchronously"""
# Handle baseline setup
sup = Supervisor(self.db)
self.sup = sup
TaskRunnerClass = MockBlueprint.TaskRunnerClass
args = MockBlueprint.ArgsClass()
args.timeout_time = 5
config = OmegaConf.structured(MephistoConfig(blueprint=args))
task_runner = TaskRunnerClass(self.task_run, config, EMPTY_STATE)
sup.register_job(self.architect, task_runner, self.provider)
self.assertEqual(len(sup.channels), 1)
channel_info = list(sup.channels.values())[0]
self.assertIsNotNone(channel_info)
self.assertTrue(channel_info.channel.is_alive())
channel_id = channel_info.channel_id
task_runner = channel_info.job.task_runner
self.assertIsNotNone(channel_id)
self.assertEqual(
len(self.architect.server.subs),
1,
"MockServer doesn't see registered channel",
)
self.assertIsNotNone(
self.architect.server.last_alive_packet,
"No alive packet received by server",
)
sup.launch_sending_thread()
self.assertIsNotNone(sup.sending_thread)
# Register a worker
mock_worker_name = "MOCK_WORKER"
self.architect.server.register_mock_worker(mock_worker_name)
workers = self.db.find_workers(worker_name=mock_worker_name)
self.assertEqual(len(workers), 1, "Worker not successfully registered")
worker = workers[0]
self.architect.server.register_mock_worker(mock_worker_name)
workers = self.db.find_workers(worker_name=mock_worker_name)
self.assertEqual(len(workers), 1, "Worker potentially re-registered")
worker_id = workers[0].db_id
self.assertEqual(len(task_runner.running_assignments), 0)
# Register an agent
mock_agent_details = "FAKE_ASSIGNMENT"
self.architect.server.register_mock_agent(worker_id,
mock_agent_details)
agents = self.db.find_agents()
self.assertEqual(len(agents), 1, "Agent was not created properly")
self.architect.server.register_mock_agent(worker_id,
mock_agent_details)
agents = self.db.find_agents()
self.assertEqual(len(agents), 1, "Agent may have been duplicated")
agent = agents[0]
self.assertIsNotNone(agent)
self.assertEqual(len(sup.agents), 1,
"Agent not registered with supervisor")
self.assertEqual(len(task_runner.running_assignments), 0,
"Task was not yet ready")
# Register another worker
mock_worker_name = "MOCK_WORKER_2"
self.architect.server.register_mock_worker(mock_worker_name)
workers = self.db.find_workers(worker_name=mock_worker_name)
worker_id = workers[0].db_id
# Register an agent
mock_agent_details = "FAKE_ASSIGNMENT_2"
self.architect.server.register_mock_agent(worker_id,
mock_agent_details)
self.assertEqual(len(task_runner.running_assignments), 1,
"Task was not launched")
agents = [a.agent for a in sup.agents.values()]
# Make both agents act
agent_id_1, agent_id_2 = agents[0].db_id, agents[1].db_id
agent_1_data = agents[0].datastore.agent_data[agent_id_1]
agent_2_data = agents[1].datastore.agent_data[agent_id_2]
self.architect.server.send_agent_act(agent_id_1, {"text": "message1"})
self.architect.server.send_agent_act(agent_id_2, {"text": "message2"})
# Give up to 1 seconds for the actual operation to occur
start_time = time.time()
TIMEOUT_TIME = 1
while time.time() - start_time < TIMEOUT_TIME:
if len(agent_1_data["acts"]) > 0:
break
time.sleep(0.1)
self.assertLess(time.time() - start_time, TIMEOUT_TIME,
"Did not process messages in time")
# Give up to 1 seconds for the task to complete afterwards
start_time = time.time()
TIMEOUT_TIME = 1
while time.time() - start_time < TIMEOUT_TIME:
if len(task_runner.running_assignments) == 0:
break
time.sleep(0.1)
self.assertLess(time.time() - start_time, TIMEOUT_TIME,
"Did not complete task in time")
# Give up to 1 seconds for all messages to propogate
start_time = time.time()
TIMEOUT_TIME = 1
while time.time() - start_time < TIMEOUT_TIME:
if self.architect.server.actions_observed == 2:
break
time.sleep(0.1)
self.assertLess(time.time() - start_time, TIMEOUT_TIME,
"Not all actions observed in time")
sup.shutdown()
self.assertTrue(channel_info.channel.is_closed())
def test_register_concurrent_job_with_onboarding(self):
"""Test registering and running a job with onboarding"""
# Handle baseline setup
sup = Supervisor(self.db)
self.sup = sup
TEST_QUALIFICATION_NAME = "test_onboarding_qualification"
task_run_args = self.task_run.args
task_run_args.blueprint.use_onboarding = True
task_run_args.blueprint.onboarding_qualification = TEST_QUALIFICATION_NAME
task_run_args.blueprint.timeout_time = 5
task_run_args.blueprint.is_concurrent = True
self.task_run.get_task_config()
# Supervisor expects that blueprint setup has already occurred
blueprint = self.task_run.get_blueprint()
TaskRunnerClass = MockBlueprint.TaskRunnerClass
task_runner = TaskRunnerClass(self.task_run, task_run_args,
EMPTY_STATE)
sup.register_job(self.architect, task_runner, self.provider)
self.assertEqual(len(sup.channels), 1)
channel_info = list(sup.channels.values())[0]
self.assertIsNotNone(channel_info)
self.assertTrue(channel_info.channel.is_alive())
channel_id = channel_info.channel_id
task_runner = channel_info.job.task_runner
self.assertIsNotNone(channel_id)
self.assertEqual(
len(self.architect.server.subs),
1,
"MockServer doesn't see registered channel",
)
self.assertIsNotNone(
self.architect.server.last_alive_packet,
"No alive packet received by server",
)
sup.launch_sending_thread()
self.assertIsNotNone(sup.sending_thread)
self.assertEqual(len(task_runner.running_units), 0)
# Fail to register an agent who fails onboarding
mock_worker_name = "BAD_WORKER"
self.architect.server.register_mock_worker(mock_worker_name)
workers = self.db.find_workers(worker_name=mock_worker_name)
self.assertEqual(len(workers), 1, "Worker not successfully registered")
worker_0 = workers[0]
self.architect.server.register_mock_worker(mock_worker_name)
workers = self.db.find_workers(worker_name=mock_worker_name)
self.assertEqual(len(workers), 1, "Worker potentially re-registered")
worker_id = workers[0].db_id
mock_agent_details = "FAKE_ASSIGNMENT"
self.architect.server.register_mock_agent(worker_id,
mock_agent_details)
agents = self.db.find_agents()
self.assertEqual(len(agents), 0,
"Agent should not be created yet - need onboarding")
onboard_agents = self.db.find_onboarding_agents()
self.assertEqual(len(onboard_agents), 1,
"Onboarding agent should have been created")
time.sleep(0.1)
last_packet = self.architect.server.last_packet
self.assertIsNotNone(last_packet)
self.assertIn("onboard_data", last_packet["data"],
"Onboarding not triggered")
self.architect.server.last_packet = None
# Submit onboarding from the agent
onboard_data = {"should_pass": False}
self.architect.server.register_mock_agent_after_onboarding(
worker_id, onboard_agents[0].get_agent_id(), onboard_data)
agents = self.db.find_agents()
self.assertEqual(len(agents), 0,
"Failed agent created after onboarding")
# Re-register as if refreshing
self.architect.server.register_mock_agent(worker_id,
mock_agent_details)
agents = self.db.find_agents()
self.assertEqual(len(agents), 0,
"Failed agent created after onboarding")
self.assertEqual(len(sup.agents), 0,
"Failed agent registered with supervisor")
self.assertEqual(
len(task_runner.running_units),
0,
"Task should not launch with failed worker",
)
# Register a worker
mock_worker_name = "MOCK_WORKER"
self.architect.server.register_mock_worker(mock_worker_name)
workers = self.db.find_workers(worker_name=mock_worker_name)
self.assertEqual(len(workers), 1, "Worker not successfully registered")
worker_1 = workers[0]
self.architect.server.register_mock_worker(mock_worker_name)
workers = self.db.find_workers(worker_name=mock_worker_name)
self.assertEqual(len(workers), 1, "Worker potentially re-registered")
worker_id = workers[0].db_id
self.assertEqual(len(task_runner.running_assignments), 0)
# Fail to register a blocked agent
mock_agent_details = "FAKE_ASSIGNMENT"
qualification_id = blueprint.onboarding_qualification_id
self.db.grant_qualification(qualification_id, worker_1.db_id, 0)
self.architect.server.register_mock_agent(worker_id,
mock_agent_details)
agents = self.db.find_agents()
self.assertEqual(len(agents), 0,
"Agent should not be created yet, failed onboarding")
time.sleep(0.1)
last_packet = self.architect.server.last_packet
self.assertIsNotNone(last_packet)
self.assertNotIn(
"onboard_data",
last_packet["data"],
"Onboarding triggered for disqualified worker",
)
self.assertIsNone(last_packet["data"]["agent_id"],
"worker assigned real agent id")
self.architect.server.last_packet = None
self.db.revoke_qualification(qualification_id, worker_id)
# Register an onboarding agent successfully
mock_agent_details = "FAKE_ASSIGNMENT"
self.architect.server.register_mock_agent(worker_id,
mock_agent_details)
agents = self.db.find_agents()
self.assertEqual(len(agents), 0,
"Agent should not be created yet - need onboarding")
onboard_agents = self.db.find_onboarding_agents()
self.assertEqual(len(onboard_agents), 2,
"Onboarding agent should have been created")
time.sleep(0.1)
last_packet = self.architect.server.last_packet
self.assertIsNotNone(last_packet)
self.assertIn("onboard_data", last_packet["data"],
"Onboarding not triggered")
self.architect.server.last_packet = None
# Submit onboarding from the agent
onboard_data = {"should_pass": True}
self.architect.server.register_mock_agent_after_onboarding(
worker_id, onboard_agents[1].get_agent_id(), onboard_data)
agents = self.db.find_agents()
self.assertEqual(len(agents), 1, "Agent not created after onboarding")
# Re-register as if refreshing
self.architect.server.register_mock_agent(worker_id,
mock_agent_details)
agents = self.db.find_agents()
self.assertEqual(len(agents), 1, "Agent may have been duplicated")
agent = agents[0]
self.assertIsNotNone(agent)
self.assertEqual(len(sup.agents), 1,
"Agent not registered with supervisor")
self.assertEqual(
len(task_runner.running_assignments),
0,
"Task was not yet ready, should not launch",
)
# Register another worker
mock_worker_name = "MOCK_WORKER_2"
self.architect.server.register_mock_worker(mock_worker_name)
workers = self.db.find_workers(worker_name=mock_worker_name)
worker_2 = workers[0]
worker_id = worker_2.db_id
# Register an agent that is already qualified
mock_agent_details = "FAKE_ASSIGNMENT_2"
self.db.grant_qualification(qualification_id, worker_2.db_id, 1)
self.architect.server.register_mock_agent(worker_id,
mock_agent_details)
time.sleep(0.1)
last_packet = self.architect.server.last_packet
self.assertIsNotNone(last_packet)
self.assertNotIn(
"onboard_data",
last_packet["data"],
"Onboarding triggered for qualified agent",
)
agents = self.db.find_agents()
self.assertEqual(len(agents), 2,
"Second agent not created without onboarding")
self.assertEqual(len(task_runner.running_assignments), 1,
"Task was not launched")
self.assertFalse(worker_0.is_qualified(TEST_QUALIFICATION_NAME))
self.assertTrue(worker_0.is_disqualified(TEST_QUALIFICATION_NAME))
self.assertTrue(worker_1.is_qualified(TEST_QUALIFICATION_NAME))
self.assertFalse(worker_1.is_disqualified(TEST_QUALIFICATION_NAME))
self.assertTrue(worker_2.is_qualified(TEST_QUALIFICATION_NAME))
self.assertFalse(worker_2.is_disqualified(TEST_QUALIFICATION_NAME))
agents = [a.agent for a in sup.agents.values()]
# Make both agents act
agent_id_1, agent_id_2 = agents[0].db_id, agents[1].db_id
agent_1_data = agents[0].datastore.agent_data[agent_id_1]
agent_2_data = agents[1].datastore.agent_data[agent_id_2]
self.architect.server.send_agent_act(agent_id_1, {"text": "message1"})
self.architect.server.send_agent_act(agent_id_2, {"text": "message2"})
# Give up to 1 seconds for the actual operation to occur
start_time = time.time()
TIMEOUT_TIME = 1
while time.time() - start_time < TIMEOUT_TIME:
if len(agent_1_data["acts"]) > 0:
break
time.sleep(0.1)
self.assertLess(time.time() - start_time, TIMEOUT_TIME,
"Did not process messages in time")
# Give up to 1 seconds for the task to complete afterwards
start_time = time.time()
TIMEOUT_TIME = 1
while time.time() - start_time < TIMEOUT_TIME:
if len(task_runner.running_assignments) == 0:
break
time.sleep(0.1)
self.assertLess(time.time() - start_time, TIMEOUT_TIME,
"Did not complete task in time")
# Give up to 1 seconds for all messages to propogate
start_time = time.time()
TIMEOUT_TIME = 1
while time.time() - start_time < TIMEOUT_TIME:
if self.architect.server.actions_observed == 2:
break
time.sleep(0.1)
self.assertLess(time.time() - start_time, TIMEOUT_TIME,
"Not all actions observed in time")
sup.shutdown()
self.assertTrue(channel_info.channel.is_closed())
def test_register_job_with_onboarding(self):
"""Test registering and running a job with onboarding"""
# Handle baseline setup
sup = Supervisor(self.db)
self.sup = sup
TEST_QUALIFICATION_NAME = "test_onboarding_qualification"
# Register onboarding arguments for blueprint
task_run_args = self.task_run.args
task_run_args.blueprint.use_onboarding = True
task_run_args.blueprint.onboarding_qualification = TEST_QUALIFICATION_NAME
task_run_args.blueprint.timeout_time = 5
task_run_args.blueprint.is_concurrent = False
self.task_run.get_task_config()
# Supervisor expects that blueprint setup has already occurred
blueprint = self.task_run.get_blueprint()
TaskRunnerClass = MockBlueprint.TaskRunnerClass
task_runner = TaskRunnerClass(self.task_run, task_run_args,
EMPTY_STATE)
sup.register_job(self.architect, task_runner, self.provider)
self.assertEqual(len(sup.channels), 1)
channel_info = list(sup.channels.values())[0]
self.assertIsNotNone(channel_info)
self.assertTrue(channel_info.channel.is_alive())
channel_id = channel_info.channel_id
task_runner = channel_info.job.task_runner
self.assertIsNotNone(channel_id)
self.assertEqual(
len(self.architect.server.subs),
1,
"MockServer doesn't see registered channel",
)
self.assertIsNotNone(
self.architect.server.last_alive_packet,
"No alive packet received by server",
)
sup.launch_sending_thread()
self.assertIsNotNone(sup.sending_thread)
# Register a worker
mock_worker_name = "MOCK_WORKER"
self.architect.server.register_mock_worker(mock_worker_name)
workers = self.db.find_workers(worker_name=mock_worker_name)
self.assertEqual(len(workers), 1, "Worker not successfully registered")
worker_1 = workers[0]
self.architect.server.register_mock_worker(mock_worker_name)
workers = self.db.find_workers(worker_name=mock_worker_name)
self.assertEqual(len(workers), 1, "Worker potentially re-registered")
worker_id = workers[0].db_id
self.assertEqual(len(task_runner.running_units), 0)
# Fail to register a blocked agent
mock_agent_details = "FAKE_ASSIGNMENT"
qualification_id = blueprint.onboarding_qualification_id
self.db.grant_qualification(qualification_id, worker_1.db_id, 0)
self.architect.server.register_mock_agent(worker_id,
mock_agent_details)
agents = self.db.find_agents()
self.assertEqual(len(agents), 0,
"Agent should not be created yet, failed onboarding")
time.sleep(0.1)
last_packet = self.architect.server.last_packet
self.assertIsNotNone(last_packet)
self.assertNotIn(
"onboard_data",
last_packet["data"],
"Onboarding triggered for disqualified worker",
)
self.assertIsNone(last_packet["data"]["agent_id"],
"worker assigned real agent id")
self.architect.server.last_packet = None
self.db.revoke_qualification(qualification_id, worker_id)
# Register an agent successfully
mock_agent_details = "FAKE_ASSIGNMENT"
self.architect.server.register_mock_agent(worker_id,
mock_agent_details)
agents = self.db.find_agents()
self.assertEqual(len(agents), 0,
"Agent should not be created yet - need onboarding")
onboard_agents = self.db.find_onboarding_agents()
self.assertEqual(len(onboard_agents), 1,
"Onboarding agent should have been created")
time.sleep(0.1)
last_packet = self.architect.server.last_packet
self.assertIsNotNone(last_packet)
self.assertIn("onboard_data", last_packet["data"],
"Onboarding not triggered")
self.architect.server.last_packet = None
# Submit onboarding from the agent
onboard_data = {"should_pass": False}
self.architect.server.register_mock_agent_after_onboarding(
worker_id, onboard_agents[0].get_agent_id(), onboard_data)
agents = self.db.find_agents()
self.assertEqual(len(agents), 0,
"Failed agent created after onboarding")
# Re-register as if refreshing
self.architect.server.register_mock_agent(worker_id,
mock_agent_details)
agents = self.db.find_agents()
self.assertEqual(len(agents), 0,
"Failed agent created after onboarding")
self.assertEqual(len(sup.agents), 0,
"Failed agent registered with supervisor")
self.assertEqual(
len(task_runner.running_units),
0,
"Task should not launch with failed worker",
)
# Register another worker
mock_worker_name = "MOCK_WORKER_2"
self.architect.server.register_mock_worker(mock_worker_name)
workers = self.db.find_workers(worker_name=mock_worker_name)
worker_2 = workers[0]
worker_id = worker_2.db_id
# Register an agent that is already qualified
mock_agent_details = "FAKE_ASSIGNMENT_2"
self.db.grant_qualification(qualification_id, worker_2.db_id, 1)
self.architect.server.register_mock_agent(worker_id,
mock_agent_details)
time.sleep(0.1)
last_packet = self.architect.server.last_packet
self.assertIsNotNone(last_packet)
self.assertNotIn(
"onboard_data",
last_packet["data"],
"Onboarding triggered for qualified agent",
)
agents = self.db.find_agents()
self.assertEqual(len(agents), 1,
"Second agent not created without onboarding")
self.assertEqual(len(task_runner.running_units), 1,
"Tasks were not launched")
self.assertFalse(worker_1.is_qualified(TEST_QUALIFICATION_NAME))
self.assertTrue(worker_1.is_disqualified(TEST_QUALIFICATION_NAME))
self.assertTrue(worker_2.is_qualified(TEST_QUALIFICATION_NAME))
self.assertFalse(worker_2.is_disqualified(TEST_QUALIFICATION_NAME))
# Register another worker
mock_worker_name = "MOCK_WORKER_3"
self.architect.server.register_mock_worker(mock_worker_name)
workers = self.db.find_workers(worker_name=mock_worker_name)
worker_3 = workers[0]
worker_id = worker_3.db_id
mock_agent_details = "FAKE_ASSIGNMENT_3"
self.architect.server.register_mock_agent(worker_id,
mock_agent_details)
agents = self.db.find_agents()
self.assertEqual(len(agents), 1,
"Agent should not be created yet - need onboarding")
onboard_agents = self.db.find_onboarding_agents()
self.assertEqual(len(onboard_agents), 2,
"Onboarding agent should have been created")
time.sleep(0.1)
last_packet = self.architect.server.last_packet
self.assertIsNotNone(last_packet)
self.assertIn("onboard_data", last_packet["data"],
"Onboarding not triggered")
self.architect.server.last_packet = None
# Submit onboarding from the agent
onboard_data = {"should_pass": True}
self.architect.server.register_mock_agent_after_onboarding(
worker_id, onboard_agents[1].get_agent_id(), onboard_data)
agents = self.db.find_agents()
self.assertEqual(len(agents), 2, "Agent not created after onboarding")
# Re-register as if refreshing
self.architect.server.register_mock_agent(worker_id,
mock_agent_details)
agents = self.db.find_agents()
self.assertEqual(len(agents), 2,
"Duplicate agent created after onboarding")
agent = agents[1]
self.assertIsNotNone(agent)
self.assertEqual(len(sup.agents), 2,
"Agent not registered supervisor after onboarding")
self.assertEqual(len(task_runner.running_units), 2,
"Task not launched after onboarding")
agents = [a.agent for a in sup.agents.values()]
# Make both agents act
agent_id_1, agent_id_2 = agents[0].db_id, agents[1].db_id
agent_1_data = agents[0].datastore.agent_data[agent_id_1]
agent_2_data = agents[1].datastore.agent_data[agent_id_2]
self.architect.server.send_agent_act(agent_id_1, {"text": "message1"})
self.architect.server.send_agent_act(agent_id_2, {"text": "message2"})
# Give up to 1 seconds for the actual operation to occur
start_time = time.time()
TIMEOUT_TIME = 1
while time.time() - start_time < TIMEOUT_TIME:
if len(agent_1_data["acts"]) > 0:
break
time.sleep(0.1)
self.assertLess(time.time() - start_time, TIMEOUT_TIME,
"Did not process messages in time")
# Give up to 1 seconds for the task to complete afterwards
start_time = time.time()
TIMEOUT_TIME = 1
while time.time() - start_time < TIMEOUT_TIME:
if len(task_runner.running_units) == 0:
break
time.sleep(0.1)
self.assertLess(time.time() - start_time, TIMEOUT_TIME,
"Did not complete task in time")
# Give up to 1 seconds for all messages to propogate
start_time = time.time()
TIMEOUT_TIME = 1
while time.time() - start_time < TIMEOUT_TIME:
if self.architect.server.actions_observed == 2:
break
time.sleep(0.1)
self.assertLess(time.time() - start_time, TIMEOUT_TIME,
"Not all actions observed in time")
sup.shutdown()
self.assertTrue(channel_info.channel.is_closed())