def check():
"""Checks that mephisto is setup correctly"""
from mephisto.core.local_database import LocalMephistoDB
from mephisto.core.utils import get_mock_requester
try:
db = LocalMephistoDB()
get_mock_requester(db)
except Exception as e:
click.echo("Something went wrong.")
click.echo(e)
return
click.echo("Mephisto seems to be set up correctly.")
def parse_launch_arguments(
self, args=None) -> Tuple[str, str, "MephistoDB", Dict[str, Any]]:
"""
Parse common arguments out from the command line, returns a
tuple of the architect type, the requester name to use, the
MephistoDB to run with, and any additional arguments parsed
out by the argument parser
Defaults to a mock architect with a mock requester with no arguments
"""
args, _unknown = self.parse_known_args(args=args)
arg_dict = vars(args)
requester_name = arg_dict["requester_name"]
provider_type = arg_dict["provider_type"]
architect_type = arg_dict["architect_type"]
datapath = arg_dict["datapath"]
if datapath is None:
datapath = get_root_data_dir()
database_path = os.path.join(datapath, "database.db")
db = LocalMephistoDB(database_path=database_path)
if requester_name is None:
if provider_type is None:
print("No requester specified, defaulting to mock")
provider_type = "mock"
if provider_type == "mock":
req = get_mock_requester(db)
requester_name = req.requester_name
else:
reqs = db.find_requesters(provider_type=provider_type)
# TODO (#93) proper logging
if len(reqs) == 0:
print(
f"No requesters found for provider type {provider_type}, please "
f"register one. You can register with `mephisto register {provider_type}`, "
f"or `python mephisto/client/cli.py register {provider_type}` if you haven't "
"installed Mephisto using poetry.")
exit(1)
elif len(reqs) == 1:
req = reqs[0]
requester_name = req.requester_name
print(
f"Found one `{provider_type}` requester to launch with: {requester_name}"
)
else:
req = reqs[-1]
requester_name = req.requester_name
print(
f"Found many `{provider_type}` requesters to launch with, "
f"choosing the most recent: {requester_name}")
else:
# Ensure provided requester exists
reqs = db.find_requesters(requester_name=requester_name)
if len(reqs) == 0:
print(f"No requesters found under name {requester_name}, "
"have you registered with `mephisto register`?")
exit(1)
provider_type = reqs[0].provider_type
# provider type and requester name now set, ensure architect
if architect_type is None:
if provider_type == "mock":
architect_type = "local"
elif provider_type == "mturk_sandbox":
architect_type = "heroku"
elif provider_type == "mturk":
architect_type = "heroku"
else:
architect_type = "local"
# TODO (#93) proper logging
print(f"No architect specified, defaulting to architect "
f"`{architect_type}` for provider `{provider_type}`")
if provider_type in ["mturk"]:
input(
f"This task is going to launch live on {provider_type}, press enter to continue: "
)
return architect_type, requester_name, db, arg_dict
def augment_config_from_db(script_cfg: DictConfig,
db: "MephistoDB") -> DictConfig:
"""
Check the database for validity of the incoming MephistoConfig, ensure
that the config has all the necessary fields set.
"""
cfg = script_cfg.mephisto
requester_name = cfg.provider.get("requester_name", None)
provider_type = cfg.provider.get("_provider_type", None)
architect_type = cfg.architect.get("_architect_type", None)
if requester_name is None:
if provider_type is None:
print("No requester specified, defaulting to mock")
provider_type = "mock"
if provider_type == "mock":
req = get_mock_requester(db)
requester_name = req.requester_name
else:
reqs = db.find_requesters(provider_type=provider_type)
# TODO (#93) proper logging
if len(reqs) == 0:
print(
f"No requesters found for provider type {provider_type}, please "
f"register one. You can register with `mephisto register {provider_type}`, "
f"or `python mephisto/client/cli.py register {provider_type}` if you haven't "
"installed Mephisto using poetry.")
exit(1)
elif len(reqs) == 1:
req = reqs[0]
requester_name = req.requester_name
print(
f"Found one `{provider_type}` requester to launch with: {requester_name}"
)
else:
req = reqs[-1]
requester_name = req.requester_name
print(
f"Found many `{provider_type}` requesters to launch with, "
f"choosing the most recent: {requester_name}")
else:
# Ensure provided requester exists
reqs = db.find_requesters(requester_name=requester_name)
if len(reqs) == 0:
print(f"No requesters found under name {requester_name}, "
"have you registered with `mephisto register`?")
exit(1)
provider_type = reqs[0].provider_type
if provider_type in ["mturk"]:
input(
f"This task is going to launch live on {provider_type}, press enter to continue: "
)
if provider_type in ["mturk_sandbox", "mturk"
] and architect_type != "heroku":
input(
f"This task is going to launch live on {provider_type}, but your "
f"provided architect is {architect_type}, are you sure you "
"want to do this? : ")
cfg.provider.requester_name = requester_name
cfg.provider._provider_type = provider_type
return script_cfg
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