def load_dag(self):
"""
Generates a dag (a directed acyclic execution graph).
Assigns it to `self.dag`.
"""
environment = self.expanded_spec.get_study_environment()
steps = self.expanded_spec.get_study_steps()
parameters = self.expanded_spec.get_parameters()
# Setup the study.
study = Study(
self.expanded_spec.name,
self.expanded_spec.description,
studyenv=environment,
parameters=parameters,
steps=steps,
out_path=self.workspace,
)
# Prepare the maestro study
if self.restart_dir is None:
study.setup_workspace()
study.setup_environment()
study.configure_study(
throttle=0,
submission_attempts=1,
restart_limit=0,
use_tmp=None,
hash_ws=None,
)
# Generate the DAG
_, maestro_dag = study.stage()
labels = []
if self.expanded_spec.merlin["samples"]:
labels = self.expanded_spec.merlin["samples"]["column_labels"]
self.dag = DAG(maestro_dag, labels)
def run_study(args):
"""Run a Maestro study."""
# Load the Specification
try:
spec = YAMLSpecification.load_specification(args.specification)
except jsonschema.ValidationError as e:
LOGGER.error(e.message)
sys.exit(1)
environment = spec.get_study_environment()
steps = spec.get_study_steps()
# Set up the output directory.
out_dir = environment.remove("OUTPUT_PATH")
if args.out:
# If out is specified in the args, ignore OUTPUT_PATH.
output_path = os.path.abspath(args.out)
# If we are automatically launching, just set the input as yes.
if os.path.exists(output_path):
if args.autoyes:
uinput = "y"
elif args.autono:
uinput = "n"
else:
uinput = six.moves.input(
"Output path already exists. Would you like to overwrite "
"it? [yn] ")
if uinput.lower() in ACCEPTED_INPUT:
print("Cleaning up existing out path...")
shutil.rmtree(output_path)
else:
print("Opting to quit -- not cleaning up old out path.")
sys.exit(0)
else:
if out_dir is None:
# If we don't find OUTPUT_PATH in the environment, assume pwd.
out_dir = os.path.abspath("./")
else:
# We just take the value from the environment.
out_dir = os.path.abspath(out_dir.value)
out_name = "{}_{}".format(spec.name.replace(" ", "_"),
time.strftime("%Y%m%d-%H%M%S"))
output_path = make_safe_path(out_dir, *[out_name])
environment.add(Variable("OUTPUT_PATH", output_path))
# Set up file logging
create_parentdir(os.path.join(output_path, "logs"))
log_path = os.path.join(output_path, "logs", "{}.log".format(spec.name))
LOG_UTIL.add_file_handler(log_path, LFORMAT, args.debug_lvl)
# Check for pargs without the matching pgen
if args.pargs and not args.pgen:
msg = "Cannot use the 'pargs' parameter without specifying a 'pgen'!"
LOGGER.exception(msg)
raise ArgumentError(msg)
# Addition of the $(SPECROOT) to the environment.
spec_root = os.path.split(args.specification)[0]
spec_root = Variable("SPECROOT", os.path.abspath(spec_root))
environment.add(spec_root)
# Handle loading a custom ParameterGenerator if specified.
if args.pgen:
# 'pgen_args' has a default of an empty list, which should translate
# to an empty dictionary.
kwargs = create_dictionary(args.pargs)
# Copy the Python file used to generate parameters.
shutil.copy(args.pgen, output_path)
# Add keywords and environment from the spec to pgen args.
kwargs["OUTPUT_PATH"] = output_path
kwargs["SPECROOT"] = spec_root
# Load the parameter generator.
parameters = load_parameter_generator(args.pgen, environment, kwargs)
else:
parameters = spec.get_parameters()
# Setup the study.
study = Study(spec.name,
spec.description,
studyenv=environment,
parameters=parameters,
steps=steps,
out_path=output_path)
# Check if the submission attempts is greater than 0:
if args.attempts < 1:
_msg = "Submission attempts must be greater than 0. " \
"'{}' provided.".format(args.attempts)
LOGGER.error(_msg)
raise ArgumentError(_msg)
# Check if the throttle is zero or greater:
if args.throttle < 0:
_msg = "Submission throttle must be a value of zero or greater. " \
"'{}' provided.".format(args.throttle)
LOGGER.error(_msg)
raise ArgumentError(_msg)
# Check if the restart limit is zero or greater:
if args.rlimit < 0:
_msg = "Restart limit must be a value of zero or greater. " \
"'{}' provided.".format(args.rlimit)
LOGGER.error(_msg)
raise ArgumentError(_msg)
# Set up the study workspace and configure it for execution.
study.setup_workspace()
study.configure_study(throttle=args.throttle,
submission_attempts=args.attempts,
restart_limit=args.rlimit,
use_tmp=args.usetmp,
hash_ws=args.hashws,
dry_run=args.dry)
study.setup_environment()
if args.dry:
# If performing a dry run, drive sleep time down to generate scripts.
sleeptime = 1
else:
# else, use args to decide sleeptime
sleeptime = args.sleeptime
batch = {"type": "local"}
if spec.batch:
batch = spec.batch
if "type" not in batch:
batch["type"] = "local"
# Copy the spec to the output directory
shutil.copy(args.specification, study.output_path)
# Use the Conductor's classmethod to store the study.
Conductor.store_study(study)
Conductor.store_batch(study.output_path, batch)
# If we are automatically launching, just set the input as yes.
if args.autoyes or args.dry:
uinput = "y"
elif args.autono:
uinput = "n"
else:
uinput = six.moves.input("Would you like to launch the study? [yn] ")
if uinput.lower() in ACCEPTED_INPUT:
if args.fg:
# Launch in the foreground.
LOGGER.info("Running Maestro Conductor in the foreground.")
conductor = Conductor(study)
conductor.initialize(batch, sleeptime)
completion_status = conductor.monitor_study()
conductor.cleanup()
return completion_status.value
else:
# Launch manager with nohup
log_path = make_safe_path(study.output_path,
*["{}.txt".format(study.name)])
cmd = [
"nohup", "conductor", "-t",
str(sleeptime), "-d",
str(args.debug_lvl), study.output_path, ">", log_path, "2>&1"
]
LOGGER.debug(" ".join(cmd))
start_process(" ".join(cmd))
print("Study launched successfully.")
else:
print("Study launch aborted.")
return 0
def run_study(args):
"""Run a Maestro study."""
# Load the Specification
spec = YAMLSpecification.load_specification(args.specification)
environment = spec.get_study_environment()
parameters = spec.get_parameters()
steps = spec.get_study_steps()
# Set up the output directory.
out_dir = environment.remove("OUTPUT_PATH")
if args.out:
# If out is specified in the args, ignore OUTPUT_PATH.
output_path = os.path.abspath(make_safe_path(args.out))
# If we are automatically launching, just set the input as yes.
if os.path.exists(output_path):
if args.autoyes:
uinput = "y"
elif args.autono:
uinput = "n"
else:
uinput = six.moves.input(
"Output path already exists. Would you like to overwrite "
"it? [yn] ")
if uinput.lower() in ACCEPTED_INPUT:
print("Cleaning up existing out path...")
shutil.rmtree(output_path)
else:
print("Opting to quit -- not cleaning up old out path.")
sys.exit(0)
else:
if out_dir is None:
# If we don't find OUTPUT_PATH in the environment, assume pwd.
out_dir = os.path.abspath("./")
else:
# We just take the value from the environment.
out_dir = os.path.abspath(out_dir.value)
out_name = "{}_{}".format(spec.name.replace(" ", "_"),
time.strftime("%Y%m%d-%H%M%S"))
output_path = make_safe_path(out_dir, out_name)
environment.add(Variable("OUTPUT_PATH", output_path))
# Now that we know outpath, set up logging.
setup_logging(args, output_path, spec.name)
# Addition of the $(SPECROOT) to the environment.
spec_root = os.path.split(args.specification)[0]
spec_root = Variable("SPECROOT", os.path.abspath(spec_root))
environment.add(spec_root)
# Setup the study.
study = Study(spec.name,
spec.description,
studyenv=environment,
parameters=parameters,
steps=steps,
out_path=output_path)
# Check if the submission attempts is greater than 0:
if args.attempts < 1:
_msg = "Submission attempts must be greater than 0. " \
"'{}' provided.".format(args.attempts)
LOGGER.error(_msg)
raise ArgumentError(_msg)
# Check if the throttle is zero or greater:
if args.throttle < 0:
_msg = "Submission throttle must be a value of zero or greater. " \
"'{}' provided.".format(args.throttle)
LOGGER.error(_msg)
raise ArgumentError(_msg)
# Check if the restart limit is zero or greater:
if args.rlimit < 0:
_msg = "Restart limit must be a value of zero or greater. " \
"'{}' provided.".format(args.rlimit)
LOGGER.error(_msg)
raise ArgumentError(_msg)
study.setup(throttle=args.throttle,
submission_attempts=args.attempts,
restart_limit=args.rlimit,
use_tmp=args.usetmp)
# Stage the study.
path, exec_dag = study.stage()
if not spec.batch:
exec_dag.set_adapter({"type": "local"})
else:
exec_dag.set_adapter(spec.batch)
# Copy the spec to the output directory
shutil.copy(args.specification, path)
# Check for a dry run
if args.dryrun:
raise NotImplementedError("The 'dryrun' mode is in development.")
# Pickle up the DAG
pkl_path = os.path.join(path, "{}.pkl".format(study.name))
exec_dag.pickle(pkl_path)
# If we are automatically launching, just set the input as yes.
if args.autoyes:
uinput = "y"
elif args.autono:
uinput = "n"
else:
uinput = six.moves.input("Would you like to launch the study? [yn] ")
if uinput.lower() in ACCEPTED_INPUT:
if args.fg:
# Launch in the foreground.
LOGGER.info("Running Maestro Conductor in the foreground.")
cancel_path = os.path.join(path, ".cancel.lock")
monitor_study(exec_dag, pkl_path, cancel_path, args.sleeptime)
else:
# Launch manager with nohup
cmd = [
"nohup", "conductor", "-t",
str(args.sleeptime), "-d",
str(args.debug_lvl), path, "&>",
"{}.txt".format(os.path.join(study.output_path, exec_dag.name))
]
LOGGER.debug(" ".join(cmd))
Popen(" ".join(cmd), shell=True, stdout=PIPE, stderr=PIPE)
return 0
def run_study(args):
"""Run a Maestro study."""
# Load the Specification
spec = YAMLSpecification.load_specification(args.specification)
environment = spec.get_study_environment()
steps = spec.get_study_steps()
# Set up the output directory.
out_dir = environment.remove("OUTPUT_PATH")
if args.out:
# If out is specified in the args, ignore OUTPUT_PATH.
output_path = os.path.abspath(args.out)
# If we are automatically launching, just set the input as yes.
if os.path.exists(output_path):
if args.autoyes:
uinput = "y"
elif args.autono:
uinput = "n"
else:
uinput = six.moves.input(
"Output path already exists. Would you like to overwrite "
"it? [yn] ")
if uinput.lower() in ACCEPTED_INPUT:
print("Cleaning up existing out path...")
shutil.rmtree(output_path)
else:
print("Opting to quit -- not cleaning up old out path.")
sys.exit(0)
else:
if out_dir is None:
# If we don't find OUTPUT_PATH in the environment, assume pwd.
out_dir = os.path.abspath("./")
else:
# We just take the value from the environment.
out_dir = os.path.abspath(out_dir.value)
out_name = "{}_{}".format(spec.name.replace(" ", "_"),
time.strftime("%Y%m%d-%H%M%S"))
output_path = make_safe_path(out_dir, *[out_name])
environment.add(Variable("OUTPUT_PATH", output_path))
# Now that we know outpath, set up logging.
setup_logging(args, output_path, spec.name.replace(" ", "_").lower())
# Check for pargs without the matching pgen
if args.pargs and not args.pgen:
msg = "Cannot use the 'pargs' parameter without specifying a 'pgen'!"
LOGGER.exception(msg)
raise ArgumentError(msg)
# Handle loading a custom ParameterGenerator if specified.
if args.pgen:
# 'pgen_args' has a default of an empty list, which should translate
# to an empty dictionary.
kwargs = create_dictionary(args.pargs)
# Copy the Python file used to generate parameters.
shutil.copy(args.pgen, output_path)
parameters = load_parameter_generator(args.pgen, kwargs)
else:
parameters = spec.get_parameters()
# Addition of the $(SPECROOT) to the environment.
spec_root = os.path.split(args.specification)[0]
spec_root = Variable("SPECROOT", os.path.abspath(spec_root))
environment.add(spec_root)
# Setup the study.
study = Study(spec.name,
spec.description,
studyenv=environment,
parameters=parameters,
steps=steps,
out_path=output_path)
# Check if the submission attempts is greater than 0:
if args.attempts < 1:
_msg = "Submission attempts must be greater than 0. " \
"'{}' provided.".format(args.attempts)
LOGGER.error(_msg)
raise ArgumentError(_msg)
# Check if the throttle is zero or greater:
if args.throttle < 0:
_msg = "Submission throttle must be a value of zero or greater. " \
"'{}' provided.".format(args.throttle)
LOGGER.error(_msg)
raise ArgumentError(_msg)
# Check if the restart limit is zero or greater:
if args.rlimit < 0:
_msg = "Restart limit must be a value of zero or greater. " \
"'{}' provided.".format(args.rlimit)
LOGGER.error(_msg)
raise ArgumentError(_msg)
# Set up the study workspace and configure it for execution.
study.setup_workspace()
study.setup_environment()
study.configure_study(throttle=args.throttle,
submission_attempts=args.attempts,
restart_limit=args.rlimit,
use_tmp=args.usetmp,
hash_ws=args.hashws)
# Stage the study.
path, exec_dag = study.stage()
# Write metadata
study.store_metadata()
if not spec.batch:
exec_dag.set_adapter({"type": "local"})
else:
if "type" not in spec.batch:
spec.batch["type"] = "local"
exec_dag.set_adapter(spec.batch)
# Copy the spec to the output directory
shutil.copy(args.specification, path)
# Check for a dry run
if args.dryrun:
raise NotImplementedError("The 'dryrun' mode is in development.")
# Pickle up the DAG
pkl_path = make_safe_path(path, *["{}.pkl".format(study.name)])
exec_dag.pickle(pkl_path)
# If we are automatically launching, just set the input as yes.
if args.autoyes:
uinput = "y"
elif args.autono:
uinput = "n"
else:
uinput = six.moves.input("Would you like to launch the study? [yn] ")
if uinput.lower() in ACCEPTED_INPUT:
if args.fg:
# Launch in the foreground.
LOGGER.info("Running Maestro Conductor in the foreground.")
cancel_path = os.path.join(path, ".cancel.lock")
# capture the StudyStatus enum to return
completion_status = monitor_study(exec_dag, pkl_path, cancel_path,
args.sleeptime)
return completion_status.value
else:
# Launch manager with nohup
log_path = make_safe_path(study.output_path,
*["{}.txt".format(exec_dag.name)])
cmd = [
"nohup", "conductor", "-t",
str(args.sleeptime), "-d",
str(args.debug_lvl), path, "&>", log_path
]
LOGGER.debug(" ".join(cmd))
start_process(" ".join(cmd))
print("Study launched successfully.")
else:
print("Study launch aborted.")
return 0
def main():
"""
The launcher main function.
This function uses command line arguments to locate the study description.
It makes use of the maestrowf core data structures as a high level class
inerface.
"""
# Set up the necessary base data structures to begin study set up.
parser = setup_argparser()
args = parser.parse_args()
# Load the Specification
spec = YAMLSpecification.load_specification(args.specification)
environment = spec.get_study_environment()
parameters = spec.get_parameters()
steps = spec.get_study_steps()
# Addition of the $(SPECROOT) to the environment.
spec_root = os.path.split(args.specification)[0]
spec_root = Variable("SPECROOT", os.path.abspath(spec_root))
environment.add(spec_root)
# Setup the study.
study = Study(spec.name,
spec.description,
studyenv=environment,
parameters=parameters,
steps=steps)
study.setup()
setup_logging(args, study.output_path, study.name)
# Stage the study.
path, exec_dag = study.stage()
if not spec.batch:
exec_dag.set_adapter({"type": "local"})
else:
exec_dag.set_adapter(spec.batch)
# Copy the spec to the output directory
shutil.copy(args.specification, path)
# Generate scripts
exec_dag.generate_scripts()
exec_dag.pickle(os.path.join(path, "{}.pkl".format(study.name)))
# If we are automatically launching, just set the input as yes.
if args.autoyes:
uinput = "y"
else:
uinput = raw_input("Would you like to launch the study?[yn] ")
if uinput.lower() in ACCEPTED_INPUT:
# Launch manager with nohup
cmd = [
"nohup", "conductor", "-t",
str(args.sleeptime), "-d",
str(args.debug_lvl), path, "&>",
"{}.txt".format(os.path.join(study.output_path, exec_dag.name))
]
LOGGER.debug(" ".join(cmd))
Popen(" ".join(cmd), shell=True, stdout=PIPE, stderr=PIPE)
sys.exit(0)