def create_pickled_cluster_input_file(storage_path, mapper=None, aggregator=None, reducer=None, model_class=None):
from molnsutil import molns_cloudpickle
if model_class is None and mapper is None and aggregator is None and reducer is None:
return None
unpickled_list = dict(model_class=molns_cloudpickle.dumps(model_class), mapper=mapper, aggregator=aggregator,
reducer=reducer)
with open(storage_path, "wb") as input_file:
molns_cloudpickle.dump(unpickled_list, input_file)
def run_async(self, mapper=None, aggregator=None, reducer=None, number_of_trajectories=None, result_list=None,
store_realizations=True, add_realizations=False, realizations_storage_directory=None):
""" Creates a new remote_job and deploys it on the cluster. Returns RemoteJob deployed. """
# Verify that given parameters are not referenced from other modules, as that produces referenced cloudpickling.
#calling_module = inspect.getmodule(inspect.stack()[1][0])
#logging.info("Caller module: {0}".format(calling_module))
# calling_module_name = calling_module.__name__ if calling_module is not None else None
# ClusterParameterSweep.check_ingredients_to_be_pickled(self.model_cls, mapper, aggregator, reducer,
# module_name=calling_module_name)
# Create new remote job.
job_id = create_new_id()
input_file_dir = os.path.join(constants.ClusterJobsScratchDir, constants.ClusterJobFilePrefix + job_id)
if not os.path.exists(input_file_dir):
os.makedirs(input_file_dir)
# Set input data according to the operation being performed.
input_data = {'number_of_trajectories': number_of_trajectories, 'params': self.parameters,
'store_realizations': store_realizations, 'num_engines': self.num_engines,
'is_parameter_sweep': self.is_parameter_sweep}
if add_realizations is True:
input_data['add_realizations'] = True
if realizations_storage_directory is not None:
input_data["result_list"] = result_list
input_data['realizations_storage_directory'] = realizations_storage_directory
# Write job input file.
input_file_path = os.path.join(input_file_dir, constants.ClusterExecInputFile)
with open(input_file_path, "wb") as input_file:
cloudpickle.dump(input_data, input_file)
# Create pickled_cluster_input_file.
pickled_cluster_input_file = os.path.join(input_file_dir, constants.PickledClusterInputFile)
create_pickled_cluster_input_file(storage_path=pickled_cluster_input_file, mapper=mapper,
aggregator=aggregator,
model_class=self.model_cls, reducer=reducer)
remote_job = RemoteJob(input_files=[input_file_path, pickled_cluster_input_file],
is_parameter_sweep=self.is_parameter_sweep,
date=str(datetime.datetime.now()), remote_host=self.remote_host, remote_job_id=job_id,
local_scratch_dir=input_file_dir, num_engines=self.num_engines)
# Deploy remote job.
self.cluster_deploy.deploy_job_to_cluster(remote_job)
logging.info("Deployed\n{0}".format(str(remote_job)))
return remote_job
def create_pickled_cluster_input_file(storage_path,
mapper=None,
aggregator=None,
reducer=None,
model_class=None):
from molnsutil import molns_cloudpickle
if model_class is None and mapper is None and aggregator is None and reducer is None:
return None
unpickled_list = dict(model_class=molns_cloudpickle.dumps(model_class),
mapper=mapper,
aggregator=aggregator,
reducer=reducer)
with open(storage_path, "wb") as input_file:
molns_cloudpickle.dump(unpickled_list, input_file)
num_of_trajectories = unpickled_list['chunk_size']
seed = unpickled_list['seed']
params = unpickled_list['pset']
param_set_id_ = unpickled_list['pndx']
if not unpickled_list.get('model_class', False):
with open(constants.pickled_cluster_input_file, "rb") as inp:
unpickled_cluster_input = pickle.load(inp)
model_cls = unpickled_cluster_input['model_class']
mapper_fn = unpickled_cluster_input['mapper']
aggregator_fn = unpickled_cluster_input['aggregator']
else:
model_cls = unpickled_list['model_class']
mapper_fn = unpickled_list['mapper']
aggregator_fn = unpickled_list['aggregator']
result = run_ensemble_map_and_aggregate(
model_class=model_cls,
parameters=params,
param_set_id=param_set_id_,
seed_base=seed,
number_of_trajectories=num_of_trajectories,
mapper=mapper_fn,
aggregator=aggregator_fn,
cluster_import=True)
with open(constants.job_output_file_name, "wb") as output:
cloudpickle.dump(result, output)
except Exception as errors:
with open(constants.job_error_file_name, "wb") as error:
error.write(str(errors))
with open(constants.job_input_file_name, "rb") as inp:
unpickled_list = pickle.load(inp)
num_of_trajectories = unpickled_list['pchunk']
seed = unpickled_list['seed']
params = unpickled_list['pset']
param_set_id_ = unpickled_list['pndx']
storage_mode = unpickled_list['storage_mode']
if not unpickled_list.get('model_class', False):
with open(constants.pickled_cluster_input_file, "rb") as inp:
unpickled_cluster_input = pickle.load(inp)
model_cls = unpickled_cluster_input['model_class']
else:
model_cls = unpickled_list['model_class']
result = run_ensemble(model_class=model_cls,
parameters=params,
param_set_id=param_set_id_,
seed_base=seed,
number_of_trajectories=num_of_trajectories,
storage_mode=storage_mode,
local_storage_path=os.path.dirname(
os.path.abspath(__file__)),
cluster_import=True)
with open(constants.job_output_file_name, "wb") as output:
molns_cloudpickle.dump(result, output)
except Exception as errors:
with open(constants.job_run_ensemble_error_file_name, "wb") as error:
error.write(str(errors))
if __name__ == "__main__":
try:
import molnsutil.constants as constants
import molnsutil.molns_cloudpickle as molns_cloudpickle
with open(constants.job_input_file_name, "rb") as inp:
unpickled_list = pickle.load(inp)
num_of_trajectories = unpickled_list['pchunk']
seed = unpickled_list['seed']
params = unpickled_list['pset']
param_set_id_ = unpickled_list['pndx']
storage_mode = unpickled_list['storage_mode']
if not unpickled_list.get('model_class', False):
with open(constants.pickled_cluster_input_file, "rb") as inp:
unpickled_cluster_input = pickle.load(inp)
model_cls = unpickled_cluster_input['model_class']
else:
model_cls = unpickled_list['model_class']
result = run_ensemble(model_class=model_cls, parameters=params, param_set_id=param_set_id_, seed_base=seed,
number_of_trajectories=num_of_trajectories, storage_mode=storage_mode,
local_storage_path=os.path.dirname(os.path.abspath(__file__)), cluster_import=True)
with open(constants.job_output_file_name, "wb") as output:
molns_cloudpickle.dump(result, output)
except Exception as errors:
with open(constants.job_run_ensemble_error_file_name, "wb") as error:
error.write(str(errors))
import molnsutil.constants as constants
import molnsutil.molns_cloudpickle as cloudpickle
with open(constants.job_input_file_name, "rb") as inp:
unpickled_list = pickle.load(inp)
num_of_trajectories = unpickled_list['chunk_size']
seed = unpickled_list['seed']
params = unpickled_list['pset']
param_set_id_ = unpickled_list['pndx']
if not unpickled_list.get('model_class', False):
with open(constants.pickled_cluster_input_file, "rb") as inp:
unpickled_cluster_input = pickle.load(inp)
model_cls = unpickled_cluster_input['model_class']
mapper_fn = unpickled_cluster_input['mapper']
aggregator_fn = unpickled_cluster_input['aggregator']
else:
model_cls = unpickled_list['model_class']
mapper_fn = unpickled_list['mapper']
aggregator_fn = unpickled_list['aggregator']
result = run_ensemble_map_and_aggregate(model_class=model_cls, parameters=params, param_set_id=param_set_id_,
seed_base=seed, number_of_trajectories=num_of_trajectories,
mapper=mapper_fn, aggregator=aggregator_fn, cluster_import=True)
with open(constants.job_output_file_name, "wb") as output:
cloudpickle.dump(result, output)
except Exception as errors:
with open(constants.job_error_file_name, "wb") as error:
error.write(str(errors))