def test_tmgr_rp_tmgr():
os.environ['RADICAL_PILOT_DBURL'] = MLAB
os.environ['ENTK_HB_INTERVAL'] = '30'
res_dict = {
'resource': 'local.localhost',
'walltime': 40,
'cpus': 20,
}
config = {"sandbox_cleanup": False, "db_cleanup": False}
rmgr_id = ru.generate_id('test.%(item_counter)04d', ru.ID_CUSTOM)
rmgr = RPRmgr(resource_desc=res_dict, sid=rmgr_id, rts_config=config)
rmgr._validate_resource_desc()
rmgr._populate()
rmgr._submit_resource_request()
tmgr = RPTmgr(sid=rmgr_id,
pending_queue=['pendingq-1'],
completed_queue=['completedq-1'],
rmgr=rmgr,
mq_hostname=hostname,
port=port)
tmgr.start_manager()
proc = Process(target=func_for_mock_tmgr_test,
args=(hostname, port, tmgr._pending_queue[0],
tmgr._completed_queue[0]))
proc.start()
proc.join()
tmgr.terminate_manager()
rmgr._terminate_resource_request()
def test_tmgr_rp_tmgr():
res_dict = {'resource': 'local.localhost', 'walltime': 40, 'cpus': 20}
config = {"sandbox_cleanup": False, "db_cleanup": False}
rmgr_id = ru.generate_id('test', ru.ID_UNIQUE)
rmgr = RPRmgr(resource_desc=res_dict, sid=rmgr_id, rts_config=config)
rmq_conn_params = pika.ConnectionParameters(host=hostname, port=port)
rmgr._validate_resource_desc()
rmgr._populate()
rmgr._submit_resource_request()
tmgr = RPTmgr(sid=rmgr_id,
pending_queue=['pendingq-1'],
completed_queue=['completedq-1'],
rmgr=rmgr,
rmq_conn_params=rmq_conn_params)
tmgr.start_manager()
proc = mp.Process(target=func_for_mock_tmgr_test,
args=(hostname, port, tmgr._pending_queue[0],
tmgr._completed_queue[0]))
proc.start()
proc.join()
tmgr.terminate_manager()
rmgr._terminate_resource_request()
def test_tmgr_rp_tmgr():
os.environ['RADICAL_PILOT_DBURL'] = MLAB
os.environ['ENTK_HB_INTERVAL'] = '30'
res_dict = {
'resource': 'local.localhost',
'walltime': 40,
'cpus': 20,
}
config={ "sandbox_cleanup": False,"db_cleanup": False}
rmgr_id = ru.generate_id('test.%(item_counter)04d', ru.ID_CUSTOM)
rmgr = RPRmgr(resource_desc=res_dict, sid=rmgr_id, rts_config=config)
rmgr._validate_resource_desc()
rmgr._populate()
rmgr._submit_resource_request()
tmgr = RPTmgr(sid=rmgr_id,
pending_queue=['pendingq-1'],
completed_queue=['completedq-1'],
rmgr=rmgr,
mq_hostname=hostname,
port=port)
tmgr.start_manager()
proc = Process(target=func_for_mock_tmgr_test, args=(hostname,
port,
tmgr._pending_queue[0],
tmgr._completed_queue[0]))
proc.start()
proc.join()
tmgr.terminate_manager()
rmgr._terminate_resource_request()
def test_write_session_description():
amgr = AppManager(hostname=hostname, port=port)
amgr.resource_desc = {'resource' : 'xsede.stampede',
'walltime' : 59,
'cpus' : 128,
'gpus' : 64,
'project' : 'xyz',
'queue' : 'high'}
workflow = [generate_pipeline(1), generate_pipeline(2)]
amgr.workflow = workflow
amgr._wfp = WFprocessor(sid=amgr.sid,
workflow=amgr._workflow,
pending_queue=amgr._pending_queue,
completed_queue=amgr._completed_queue,
resubmit_failed=amgr._resubmit_failed,
rmq_conn_params=amgr._rmq_conn_params)
amgr._workflow = amgr._wfp.workflow
amgr._task_manager = TaskManager(sid=amgr._sid,
pending_queue=amgr._pending_queue,
completed_queue=amgr._completed_queue,
rmgr=amgr._rmgr,
rmq_conn_params=amgr._rmq_conn_params
)
write_session_description(amgr)
desc = ru.read_json('%s/radical.entk.%s.json' % (amgr._sid, amgr._sid))
src = '%s/sample_data' % pwd
assert desc == ru.read_json('%s/expected_desc_write_session.json' % src)
def test_amgr_resource_terminate():
res_dict = {
'resource': 'xsede.supermic',
'walltime': 30,
'cpus': 20,
'project': 'TG-MCB090174'
}
from radical.entk.execman.rp import TaskManager
amgr = Amgr(rts='radical.pilot', hostname=hostname, port=port)
amgr.resource_desc = res_dict
amgr._setup_mqs()
amgr._rmq_cleanup = True
amgr._task_manager = TaskManager(sid='test',
pending_queue=list(),
completed_queue=list(),
mq_hostname=amgr._mq_hostname,
rmgr=amgr._resource_manager,
port=amgr._port
)
amgr.resource_terminate()
def test_write_session_description():
amgr = AppManager(hostname=hostname,
port=port,
username=username,
password=password)
amgr.resource_desc = {
'resource': 'xsede.stampede',
'walltime': 59,
'cpus': 128,
'gpus': 64,
'project': 'xyz',
'queue': 'high'
}
workflow = [generate_pipeline(1), generate_pipeline(2)]
amgr.workflow = workflow
amgr._wfp = WFprocessor(sid=amgr.sid,
workflow=amgr._workflow,
pending_queue=amgr._pending_queue,
completed_queue=amgr._completed_queue,
resubmit_failed=amgr._resubmit_failed,
rmq_conn_params=amgr._rmq_conn_params)
amgr._workflow = amgr._wfp.workflow
amgr._task_manager = TaskManager(sid=amgr._sid,
pending_queue=amgr._pending_queue,
completed_queue=amgr._completed_queue,
rmgr=amgr._rmgr,
rmq_conn_params=amgr._rmq_conn_params)
write_session_description(amgr)
desc = ru.read_json('%s/radical.entk.%s.json' % (amgr._sid, amgr._sid))
# tasks are originally set but saved as a list in json
# uses sorting for convenient comparison, this doesn't change validity
for k, v in (desc['tree'].items()):
if k.startswith("stage"):
desc['tree'][k]['children'] = sorted(v['children'])
src = '%s/sample_data' % pwd
assert desc == ru.read_json('%s/expected_desc_write_session.json' % src)
def test_write_session_description():
hostname = os.environ.get('RMQ_HOSTNAME', 'localhost')
port = int(os.environ.get('RMQ_PORT', 5672))
amgr = AppManager(hostname=hostname, port=port)
amgr.resource_desc = {
'resource': 'xsede.stampede',
'walltime': 60,
'cpus': 128,
'gpus': 64,
'project': 'xyz',
'queue': 'high'
}
workflow = [generate_pipeline(1), generate_pipeline(2)]
amgr.workflow = workflow
amgr._wfp = WFprocessor(sid=amgr._sid,
workflow=amgr._workflow,
pending_queue=amgr._pending_queue,
completed_queue=amgr._completed_queue,
mq_hostname=amgr._mq_hostname,
port=amgr._port,
resubmit_failed=amgr._resubmit_failed)
amgr._wfp._initialize_workflow()
amgr._workflow = amgr._wfp.workflow
amgr._task_manager = TaskManager(sid=amgr._sid,
pending_queue=amgr._pending_queue,
completed_queue=amgr._completed_queue,
mq_hostname=amgr._mq_hostname,
rmgr=amgr._resource_manager,
port=amgr._port
)
# os.mkdir(amgr._sid)
write_session_description(amgr)
desc = ru.read_json('%s/radical.entk.%s.json' % (amgr._sid, amgr._sid))
curdir = os.path.dirname(os.path.abspath(__file__))
src = '%s/sample_data' % curdir
assert desc == ru.read_json('%s/expected_desc_write_session.json' % src)
def run(self):
"""
**Purpose**: Run the application manager. Once the workflow and resource manager have been assigned. Invoking this
method will start the setting up the communication infrastructure, submitting a resource request and then
submission of all the tasks.
"""
try:
# Set None objects local to each run
self._wfp = None
self._sync_thread = None
self._terminate_sync = Event()
self._resubmit_failed = False
self._cur_attempt = 1
if not self._workflow:
self._logger.error('No workflow assigned currently, please check your script')
raise MissingError(obj=self._uid, missing_attribute='workflow')
if not self._resource_manager:
self._logger.error('No resource manager assigned currently, please create and add a valid resource manager')
raise MissingError(obj=self._uid, missing_attribute='resource_manager')
self._prof.prof('amgr run started', uid=self._uid)
# Setup rabbitmq stuff
if not self._mqs_setup:
self._report.info('Setting up RabbitMQ system')
setup = self._setup_mqs()
if not setup:
self._logger.error('RabbitMQ system not available')
raise EnTKError("RabbitMQ setup failed")
self._mqs_setup = True
self._report.ok('>>ok\n')
# Create WFProcessor object
self._prof.prof('creating wfp obj', uid=self._uid)
self._wfp = WFprocessor(sid=self._sid,
workflow=self._workflow,
pending_queue=self._pending_queue,
completed_queue=self._completed_queue,
mq_hostname=self._mq_hostname,
port=self._port,
resubmit_failed=self._resubmit_failed)
self._wfp._initialize_workflow()
self._workflow = self._wfp.workflow
# Submit resource request if not resource allocation done till now or
# resubmit a new one if the old one has completed
if self._resource_manager:
res_alloc_state = self._resource_manager.get_resource_allocation_state()
if (not res_alloc_state) or (res_alloc_state in self._resource_manager.get_completed_states()):
self._logger.info('Starting resource request submission')
self._prof.prof('init rreq submission', uid=self._uid)
self._resource_manager._submit_resource_request()
res_alloc_state = self._resource_manager.get_resource_allocation_state()
if res_alloc_state in self._resource_manager.get_completed_states():
raise EnTKError(msg="Cannot proceed. Resource allocation ended up in %s"%res_alloc_state)
else:
self._logger.exception('Cannot run without resource manager, please create and assign a resource manager')
raise EnTKError(text='Missing resource manager')
# Start synchronizer thread
if not self._sync_thread:
self._logger.info('Starting synchronizer thread')
self._sync_thread = Thread(target=self._synchronizer, name='synchronizer-thread')
self._prof.prof('starting synchronizer thread', uid=self._uid)
self._sync_thread.start()
# Start WFprocessor
self._logger.info('Starting WFProcessor process from AppManager')
self._wfp.start_processor()
self._report.ok('All components created\n')
# Create tmgr object only if it does not already exist
if self._rts == 'radical.pilot':
from radical.entk.execman.rp import TaskManager
elif self._rts == 'mock':
from radical.entk.execman.mock import TaskManager
if not self._task_manager:
self._prof.prof('creating tmgr obj', uid=self._uid)
self._task_manager = TaskManager(sid=self._sid,
pending_queue=self._pending_queue,
completed_queue=self._completed_queue,
mq_hostname=self._mq_hostname,
rmgr=self._resource_manager,
port=self._port
)
self._logger.info('Starting task manager process from AppManager')
self._task_manager.start_manager()
self._task_manager.start_heartbeat()
active_pipe_count = len(self._workflow)
finished_pipe_uids = []
# We wait till all pipelines of the workflow are marked
# complete
while ((active_pipe_count > 0) and
(self._wfp.workflow_incomplete()) and
(self._resource_manager.get_resource_allocation_state() not
in self._resource_manager.get_completed_states())):
if active_pipe_count > 0:
for pipe in self._workflow:
with pipe.lock:
if (pipe.completed) and (pipe.uid not in finished_pipe_uids):
self._logger.info('Pipe %s completed' % pipe.uid)
finished_pipe_uids.append(pipe.uid)
active_pipe_count -= 1
self._logger.info('Active pipes: %s' % active_pipe_count)
if (not self._sync_thread.is_alive()) and (self._cur_attempt <= self._reattempts):
self._sync_thread = Thread(target=self._synchronizer,
name='synchronizer-thread')
self._logger.info('Restarting synchronizer thread')
self._prof.prof('restarting synchronizer', uid=self._uid)
self._sync_thread.start()
self._cur_attempt += 1
if (not self._wfp.check_processor()) and (self._cur_attempt <= self._reattempts):
"""
If WFP dies, both child threads are also cleaned out.
We simply recreate the wfp object with a copy of the workflow
in the appmanager and start the processor.
"""
self._prof.prof('recreating wfp obj', uid=self._uid)
self._wfp = WFProcessor(
sid=self._sid,
workflow=self._workflow,
pending_queue=self._pending_queue,
completed_queue=self._completed_queue,
mq_hostname=self._mq_hostname,
port=self._port,
resubmit_failed=self._resubmit_failed)
self._logger.info('Restarting WFProcessor process from AppManager')
self._wfp.start_processor()
self._cur_attempt += 1
if (not self._task_manager.check_heartbeat()) and (self._cur_attempt <= self._reattempts):
"""
If the tmgr process or heartbeat dies, we simply start a
new process using the start_manager method. We do not
need to create a new instance of the TaskManager object
itself. We stop and start a new instance of the
heartbeat thread as well.
"""
self._prof.prof('restarting tmgr process and heartbeat', uid=self._uid)
self._logger.info('Terminating heartbeat thread')
self._task_manager.terminate_heartbeat()
self._logger.info('Terminating tmgr process')
self._task_manager.terminate_manager()
self._logger.info('Restarting task manager process')
self._task_manager.start_manager()
self._logger.info('Restarting heartbeat thread')
self._task_manager.start_heartbeat()
self._cur_attempt += 1
self._prof.prof('start termination', uid=self._uid)
# Terminate threads in following order: wfp, helper, synchronizer
self._logger.info('Terminating WFprocessor')
self._wfp.terminate_processor()
self._logger.info('Terminating synchronizer thread')
self._terminate_sync.set()
self._sync_thread.join()
self._logger.info('Synchronizer thread terminated')
if self._autoterminate:
self._terminate()
if self._write_workflow:
write_workflow(self._workflow, self._sid)
self._prof.prof('termination done', uid=self._uid)
except KeyboardInterrupt:
self._prof.prof('start termination', uid=self._uid)
self._logger.exception('Execution interrupted by user (you probably hit Ctrl+C), ' +
'trying to cancel enqueuer thread gracefully...')
self._terminate()
self._prof.prof('termination done', uid=self._uid)
raise KeyboardInterrupt
except Exception, ex:
self._prof.prof('start termination', uid=self._uid)
self._logger.exception('Error in AppManager: %s' % ex)
# Terminate threads in following order: wfp, helper, synchronizer
self._terminate()
self._prof.prof('termination done', uid=self._uid)
raise
class AppManager(object):
"""
An application manager takes the responsibility of setting up the communication infrastructure, instantiates the
ResourceManager, TaskManager, WFProcessor objects and all their threads and processes. This is the Master object
running in the main process and is designed to recover from errors from all other objects, threads and processes.
:Arguments:
:config_path: Url to config path to be read for AppManager
:hostname: host rabbitmq server is running
:port: port at which rabbitmq can be accessed
:reattempts: number of attempts to re-invoke any failed EnTK components
:resubmit_failed: resubmit failed tasks (True/False)
:autoterminate: terminate resource reservation upon execution of all tasks of first workflow (True/False)
:write_workflow: write workflow and mapping to rts entities to a file (post-termination)
:rts: Specify RTS to use. Current options: 'mock', 'radical.pilot' (default if unspecified)
:rmq_cleanup: Cleanup all queues created in RabbitMQ server for current execution (default is True)
:rts_config: Configuration for the RTS, accepts {"sandbox_cleanup": True/False,"db_cleanup": True/False} when RTS is RP
:name: Name of the Application. It should be unique between executions. (default is randomly assigned)
"""
def __init__(self,
config_path=None,
hostname=None,
port=None,
reattempts=None,
resubmit_failed=None,
autoterminate=None,
write_workflow=None,
rts=None,
rmq_cleanup=None,
rts_config=None,
name=None):
# Create a session for each EnTK script execution
if name:
self._name = name
self._sid = name
else:
self._name= str()
self._sid = ru.generate_id('re.session', ru.ID_PRIVATE)
self._read_config(config_path, hostname, port, reattempts,
resubmit_failed, autoterminate, write_workflow,
rts, rmq_cleanup, rts_config)
# Create an uid + logger + profiles for AppManager, under the sid
# namespace
path = os.getcwd() + '/' + self._sid
self._uid = ru.generate_id('appmanager.%(item_counter)04d', ru.ID_CUSTOM, namespace=self._sid)
self._logger = ru.Logger('radical.entk.%s' % self._uid, path=path)
self._prof = ru.Profiler(name='radical.entk.%s' % self._uid, path=path)
self._report = ru.Reporter(name='radical.entk.%s' % self._uid)
self._report.info('EnTK session: %s\n' % self._sid)
self._prof.prof('create amgr obj', uid=self._uid)
self._report.info('Creating AppManager')
self._resource_manager = None
# RabbitMQ Queues
self._pending_queue = list()
self._completed_queue = list()
# Global parameters to have default values
self._mqs_setup = False
self._resource_desc = None
self._task_manager = None
self._workflow = None
self._cur_attempt = 1
self._shared_data = list()
self._wfp = None
self._sync_thread = None
self._rmq_ping_interval = os.getenv('RMQ_PING_INTERVAL', 10)
self._logger.info('Application Manager initialized')
self._prof.prof('amgr obj created', uid=self._uid)
self._report.ok('>>ok\n')
def _read_config(self, config_path, hostname, port, reattempts,
resubmit_failed, autoterminate, write_workflow,
rts, rmq_cleanup, rts_config):
if not config_path:
config_path = os.path.dirname(os.path.abspath(__file__))
config = ru.read_json(os.path.join(config_path, 'config.json'))
self._mq_hostname = hostname if hostname else str(config['hostname'])
self._port = int(port if port else config['port'])
self._reattempts = reattempts if reattempts else config['reattempts']
self._resubmit_failed = resubmit_failed if resubmit_failed is not None else config['resubmit_failed']
self._autoterminate = autoterminate if autoterminate is not None else config['autoterminate']
self._write_workflow = write_workflow if write_workflow is not None else config['write_workflow']
self._rts = rts if rts in ['radical.pilot', 'mock'] else str(config['rts'])
self._rmq_cleanup = rmq_cleanup if rmq_cleanup is not None else config['rmq_cleanup']
self._rts_config = rts_config if rts_config is not None else config['rts_config']
self._num_pending_qs = config['pending_qs']
self._num_completed_qs = config['completed_qs']
# ------------------------------------------------------------------------------------------------------------------
# Getter functions
# ------------------------------------------------------------------------------------------------------------------
@property
def name(self):
"""
Name for the application manager. Allows the user to setup the name of
the application manager, as well as, its session ID. This name should be
unique between different EnTK executions, otherwise it will produce an
error.
:getter: Returns the name of the application manager
:setter: Assigns the name of the application manager
:type: String
"""
return self._name
@property
def sid(self):
"""
Get the session ID of the current EnTK execution
:getter: Returns the session ID of the EnTK execution
:type: String
"""
return self._sid
@property
def resource_desc(self):
"""
:getter: Returns the resource description
:setter: Assigns a resource description
"""
return self._resource_desc
@property
def workflow(self):
"""
:getter: Return the workflow assigned for execution
:setter: Assign workflow to be executed
"""
return self._workflow
@property
def shared_data(self):
"""
:getter: Return list of filenames that are shared between multiple tasks of the application
:setter: Assign a list of names of files that need to be staged to the remote machine
"""
return self._shared_data
# ------------------------------------------------------------------------------------------------------------------
# Setter functions
# ------------------------------------------------------------------------------------------------------------------
@name.setter
def name(self, value):
if not isinstance(value, str):
raise TypeError(expected_type=str, actual_type=type(value))
else:
self._name = value
@resource_desc.setter
def resource_desc(self, value):
if self._rts == 'radical.pilot':
from radical.entk.execman.rp import ResourceManager
self._resource_manager = ResourceManager(resource_desc=value,
sid=self._sid,
rts_config=self._rts_config)
elif self._rts == 'mock':
from radical.entk.execman.mock import ResourceManager
self._resource_manager = ResourceManager(resource_desc=value,
sid=self._sid)
self._report.info('Validating and assigning resource manager')
if self._resource_manager._validate_resource_desc():
self._resource_manager._populate()
self._resource_manager.shared_data = self._shared_data
else:
self._logger.error('Could not validate resource description')
raise
self._report.ok('>>ok\n')
@workflow.setter
def workflow(self, workflow):
self._prof.prof('assigning workflow', uid=self._uid)
for p in workflow:
if not isinstance(p, Pipeline):
self._logger.info('workflow type incorrect')
raise TypeError(expected_type=['Pipeline', 'set of Pipelines'], actual_type=type(p))
p._validate()
self._workflow = workflow
self._logger.info('Workflow assigned to Application Manager')
@shared_data.setter
def shared_data(self, data):
if not isinstance(data, list):
data = [data]
for value in data:
if not isinstance(value, str):
raise TypeError(expected_type=str, actual_type=type(value))
if self._resource_manager:
self._resource_manager.shared_data = data
# ------------------------------------------------------------------------------------------------------------------
# Public methods
# ------------------------------------------------------------------------------------------------------------------
def run(self):
"""
**Purpose**: Run the application manager. Once the workflow and resource manager have been assigned. Invoking this
method will start the setting up the communication infrastructure, submitting a resource request and then
submission of all the tasks.
"""
try:
# Set None objects local to each run
self._wfp = None
self._sync_thread = None
self._terminate_sync = Event()
self._resubmit_failed = False
self._cur_attempt = 1
if not self._workflow:
self._logger.error('No workflow assigned currently, please check your script')
raise MissingError(obj=self._uid, missing_attribute='workflow')
if not self._resource_manager:
self._logger.error('No resource manager assigned currently, please create and add a valid resource manager')
raise MissingError(obj=self._uid, missing_attribute='resource_manager')
self._prof.prof('amgr run started', uid=self._uid)
# Setup rabbitmq stuff
if not self._mqs_setup:
self._report.info('Setting up RabbitMQ system')
setup = self._setup_mqs()
if not setup:
self._logger.error('RabbitMQ system not available')
raise EnTKError("RabbitMQ setup failed")
self._mqs_setup = True
self._report.ok('>>ok\n')
# Create WFProcessor object
self._prof.prof('creating wfp obj', uid=self._uid)
self._wfp = WFprocessor(sid=self._sid,
workflow=self._workflow,
pending_queue=self._pending_queue,
completed_queue=self._completed_queue,
mq_hostname=self._mq_hostname,
port=self._port,
resubmit_failed=self._resubmit_failed)
self._wfp._initialize_workflow()
self._workflow = self._wfp.workflow
# Submit resource request if not resource allocation done till now or
# resubmit a new one if the old one has completed
if self._resource_manager:
res_alloc_state = self._resource_manager.get_resource_allocation_state()
if (not res_alloc_state) or (res_alloc_state in self._resource_manager.get_completed_states()):
self._logger.info('Starting resource request submission')
self._prof.prof('init rreq submission', uid=self._uid)
self._resource_manager._submit_resource_request()
res_alloc_state = self._resource_manager.get_resource_allocation_state()
if res_alloc_state in self._resource_manager.get_completed_states():
raise EnTKError(msg="Cannot proceed. Resource allocation ended up in %s"%res_alloc_state)
else:
self._logger.exception('Cannot run without resource manager, please create and assign a resource manager')
raise EnTKError(text='Missing resource manager')
# Start synchronizer thread
if not self._sync_thread:
self._logger.info('Starting synchronizer thread')
self._sync_thread = Thread(target=self._synchronizer, name='synchronizer-thread')
self._prof.prof('starting synchronizer thread', uid=self._uid)
self._sync_thread.start()
# Start WFprocessor
self._logger.info('Starting WFProcessor process from AppManager')
self._wfp.start_processor()
self._report.ok('All components created\n')
# Create tmgr object only if it does not already exist
if self._rts == 'radical.pilot':
from radical.entk.execman.rp import TaskManager
elif self._rts == 'mock':
from radical.entk.execman.mock import TaskManager
if not self._task_manager:
self._prof.prof('creating tmgr obj', uid=self._uid)
self._task_manager = TaskManager(sid=self._sid,
pending_queue=self._pending_queue,
completed_queue=self._completed_queue,
mq_hostname=self._mq_hostname,
rmgr=self._resource_manager,
port=self._port
)
self._logger.info('Starting task manager process from AppManager')
self._task_manager.start_manager()
self._task_manager.start_heartbeat()
active_pipe_count = len(self._workflow)
finished_pipe_uids = []
# We wait till all pipelines of the workflow are marked
# complete
while ((active_pipe_count > 0) and
(self._wfp.workflow_incomplete()) and
(self._resource_manager.get_resource_allocation_state() not
in self._resource_manager.get_completed_states())):
if active_pipe_count > 0:
for pipe in self._workflow:
with pipe.lock:
if (pipe.completed) and (pipe.uid not in finished_pipe_uids):
self._logger.info('Pipe %s completed' % pipe.uid)
finished_pipe_uids.append(pipe.uid)
active_pipe_count -= 1
self._logger.info('Active pipes: %s' % active_pipe_count)
if (not self._sync_thread.is_alive()) and (self._cur_attempt <= self._reattempts):
self._sync_thread = Thread(target=self._synchronizer,
name='synchronizer-thread')
self._logger.info('Restarting synchronizer thread')
self._prof.prof('restarting synchronizer', uid=self._uid)
self._sync_thread.start()
self._cur_attempt += 1
if (not self._wfp.check_processor()) and (self._cur_attempt <= self._reattempts):
"""
If WFP dies, both child threads are also cleaned out.
We simply recreate the wfp object with a copy of the workflow
in the appmanager and start the processor.
"""
self._prof.prof('recreating wfp obj', uid=self._uid)
self._wfp = WFProcessor(
sid=self._sid,
workflow=self._workflow,
pending_queue=self._pending_queue,
completed_queue=self._completed_queue,
mq_hostname=self._mq_hostname,
port=self._port,
resubmit_failed=self._resubmit_failed)
self._logger.info('Restarting WFProcessor process from AppManager')
self._wfp.start_processor()
self._cur_attempt += 1
if (not self._task_manager.check_heartbeat()) and (self._cur_attempt <= self._reattempts):
"""
If the tmgr process or heartbeat dies, we simply start a
new process using the start_manager method. We do not
need to create a new instance of the TaskManager object
itself. We stop and start a new instance of the
heartbeat thread as well.
"""
self._prof.prof('restarting tmgr process and heartbeat', uid=self._uid)
self._logger.info('Terminating heartbeat thread')
self._task_manager.terminate_heartbeat()
self._logger.info('Terminating tmgr process')
self._task_manager.terminate_manager()
self._logger.info('Restarting task manager process')
self._task_manager.start_manager()
self._logger.info('Restarting heartbeat thread')
self._task_manager.start_heartbeat()
self._cur_attempt += 1
self._prof.prof('start termination', uid=self._uid)
# Terminate threads in following order: wfp, helper, synchronizer
self._logger.info('Terminating WFprocessor')
self._wfp.terminate_processor()
self._logger.info('Terminating synchronizer thread')
self._terminate_sync.set()
self._sync_thread.join()
self._logger.info('Synchronizer thread terminated')
if self._autoterminate:
self._terminate()
if self._write_workflow:
write_workflow(self._workflow, self._sid)
self._prof.prof('termination done', uid=self._uid)
except KeyboardInterrupt:
self._prof.prof('start termination', uid=self._uid)
self._logger.exception('Execution interrupted by user (you probably hit Ctrl+C), ' +
'trying to cancel enqueuer thread gracefully...')
self._terminate()
self._prof.prof('termination done', uid=self._uid)
raise KeyboardInterrupt
except Exception, ex:
self._prof.prof('start termination', uid=self._uid)
self._logger.exception('Error in AppManager: %s' % ex)
# Terminate threads in following order: wfp, helper, synchronizer
self._terminate()
self._prof.prof('termination done', uid=self._uid)
raise
def _start_all_comps(self):
if self._wfp:
# This condition is called when there are multiple workflows
# submitted for execution. Amgr.run() was probably called twice.
# If a WFP exists, we use the same one but with the new workflow.
# Since WFP (and its threads) and the Amgr share memory, we have
# to terminate WFP's threads, assign the new workflow and then
# start the threads again.
self._wfp.terminate_processor()
self._wfp._workflow = self._workflow
self._wfp.start_processor()
return
# Create WFProcessor and initialize workflow its contents with
# uids
self._prof.prof('wfp_create_start', uid=self._uid)
self._wfp = WFprocessor(sid=self._sid,
workflow=self._workflow,
pending_queue=self._pending_queue,
completed_queue=self._completed_queue,
resubmit_failed=self._resubmit_failed,
rmq_conn_params=self._rmq_conn_params)
self._prof.prof('wfp_create_stop', uid=self._uid)
# Start synchronizer thread AM OK
if not self._sync_thread:
self._logger.info('Starting synchronizer thread')
self._sync_thread = mt.Thread(target=self._synchronizer,
name='synchronizer-thread')
self._prof.prof('sync_thread_create', uid=self._uid)
self._sync_thread.start()
# Start WFprocessor
self._logger.info('Starting WFProcessor')
self._wfp.start_processor()
self._report.ok('All components created\n')
# Create tmgr object only if it does not already exist
if self._rts == 'radical.pilot':
from radical.entk.execman.rp import TaskManager
elif self._rts == 'mock':
from radical.entk.execman.mock import TaskManager
if not self._task_manager:
self._logger.info('Starting task manager')
self._prof.prof('tmgr_create_start', uid=self._uid)
self._task_manager = TaskManager(
sid=self._sid,
pending_queue=self._pending_queue,
completed_queue=self._completed_queue,
rmgr=self._rmgr,
rmq_conn_params=self._rmq_conn_params)
self._task_manager.start_manager()
self._task_manager.start_heartbeat()
self._prof.prof('tmgr_create_stop', uid=self._uid)
class AppManager(object):
'''
An application manager takes the responsibility of setting up the
communication infrastructure, instantiates the ResourceManager, TaskManager,
WFProcessor objects and all their threads and processes. This is the Master
object running in the main process and is designed to recover from errors
from all other objects, threads and processes.
:Arguments:
:config_path: Url to config path to be read for AppManager
:hostname: host rabbitmq server is running
:port: port at which rabbitmq can be accessed
:username: username to log in to RabbitMQ
:password: password to log in to RabbitMQ
:reattempts: number of attempts to re-invoke any failed EnTK
components
:resubmit_failed: resubmit failed tasks (True/False)
:autoterminate: terminate resource reservation upon execution of all
tasks of first workflow (True/False)
:write_workflow: write workflow and mapping to rts entities to a file
(post-termination)
:rts: Specify RTS to use. Current options: 'mock',
'radical.pilot' (default if unspecified)
:rmq_cleanup: Cleanup all queues created in RabbitMQ server for
current execution (default is True)
:rts_config: Configuration for the RTS, accepts
{'sandbox_cleanup': True/False,'db_cleanup':
True/False} when RTS is RP
:name: Name of the Application. It should be unique between
executions. (default is randomly assigned)
'''
# --------------------------------------------------------------------------
#
def __init__(self,
config_path=None,
hostname=None,
port=None,
username=None,
password=None,
reattempts=None,
resubmit_failed=None,
autoterminate=None,
write_workflow=None,
rts=None,
rmq_cleanup=None,
rts_config=None,
name=None):
# Create a session for each EnTK script execution
if name:
self._name = name
self._sid = name
else:
self._name = str()
self._sid = ru.generate_id('re.session', ru.ID_PRIVATE)
self._read_config(config_path, hostname, port, username, password,
reattempts, resubmit_failed, autoterminate,
write_workflow, rts, rmq_cleanup, rts_config)
# Create an uid + logger + profiles for AppManager, under the sid
# namespace
self._uid = ru.generate_id('appmanager.%(counter)04d', ru.ID_CUSTOM)
path = os.getcwd() + '/' + self._sid
name = 'radical.entk.%s' % self._uid
self._logger = ru.Logger(name=name, path=path)
self._prof = ru.Profiler(name=name, path=path)
self._report = ru.Reporter(name=name)
self._report.info('EnTK session: %s\n' % self._sid)
self._report.info('Creating AppManager')
self._prof.prof('amgr_creat', uid=self._uid)
self._rmgr = None
self._pending_queue = list()
self._completed_queue = list()
# Global parameters to have default values
self._mqs_setup = False
self._resource_desc = None
self._task_manager = None
self._workflow = None
self._workflows = list()
self._cur_attempt = 1
self._shared_data = list()
self._outputs = None
self._wfp = None
self._sync_thread = None
self._terminate_sync = mt.Event()
self._resubmit_failed = False
self._port = int(self._port)
# Setup rabbitmq queues
self._setup_mqs()
self._rmq_ping_interval = int(os.getenv('RMQ_PING_INTERVAL', "10"))
self._logger.info('Application Manager initialized')
self._prof.prof('amgr_created', uid=self._uid)
self._report.ok('>>ok\n')
# --------------------------------------------------------------------------
#
def _read_config(self, config_path, hostname, port, username, password,
reattempts, resubmit_failed, autoterminate,
write_workflow, rts, rmq_cleanup, rts_config):
if not config_path:
config_path = os.path.dirname(os.path.abspath(__file__))
config = ru.read_json(os.path.join(config_path, 'config.json'))
def _if(val1, val2):
if val1 is not None: return val1
else: return val2
self._hostname = _if(hostname, config['hostname'])
self._port = _if(port, config['port'])
self._username = _if(username, config['username'])
self._password = _if(password, config['password'])
self._reattempts = _if(reattempts, config['reattempts'])
self._resubmit_failed = _if(resubmit_failed, config['resubmit_failed'])
self._autoterminate = _if(autoterminate, config['autoterminate'])
self._write_workflow = _if(write_workflow, config['write_workflow'])
self._rmq_cleanup = _if(rmq_cleanup, config['rmq_cleanup'])
self._rts_config = _if(rts_config, config['rts_config'])
self._rts = _if(rts, config['rts'])
credentials = pika.PlainCredentials(self._username, self._password)
self._rmq_conn_params = pika.connection.ConnectionParameters(
host=self._hostname, port=self._port, credentials=credentials)
self._num_pending_qs = config['pending_qs']
self._num_completed_qs = config['completed_qs']
if self._rts not in ['radical.pilot', 'mock']:
raise ValueError('invalid RTS %s' % self._rts)
# --------------------------------------------------------------------------
#
# Getter functions
#
@property
def name(self):
'''
Name for the application manager. Allows the user to setup the name of
the application manager, as well as its session ID. This name should be
unique between different EnTK executions, otherwise it will produce an
error.
:getter: Returns the name of the application manager
:setter: Assigns the name of the application manager
:type: String
'''
return self._name
# --------------------------------------------------------------------------
#
@property
def sid(self):
'''
Get the session ID of the current EnTK execution
:getter: Returns the session ID of the EnTK execution
:type: String
'''
return self._sid
# --------------------------------------------------------------------------
#
@property
def resource_desc(self):
'''
:getter: Returns the resource description
:setter: Assigns a resource description
'''
return self._resource_desc
# --------------------------------------------------------------------------
#
@property
def workflow(self):
'''
:getter: Return the last workflow assigned for execution
:setter: Assign a new workflow to be executed
'''
return self._workflow
# --------------------------------------------------------------------------
#
@property
def workflows(self):
"""
:getter: Return a list of workflows assigned for execution
"""
return self._workflows
@property
def shared_data(self):
'''
:getter: Return list of filenames that are shared between multiple tasks
of the application
:setter: Assign a list of names of files that need to be staged to the
remote machine
'''
return self._shared_data
@property
def outputs(self):
'''
:getter: Return list of filenames that are to be staged out after
execution
:setter: Assign a list of names of files that need to be staged from the
remote machine
'''
return self._outputs
# --------------------------------------------------------------------------
# Setter functions
#
@name.setter
def name(self, value):
if not isinstance(value, str):
raise ree.TypeError(expected_type=str, actual_type=type(value))
self._name = value
# --------------------------------------------------------------------------
#
@resource_desc.setter
def resource_desc(self, value):
if self._rts == 'radical.pilot':
from radical.entk.execman.rp import ResourceManager
elif self._rts == 'mock':
from radical.entk.execman.mock import ResourceManager
self._rmgr = ResourceManager(resource_desc=value,
sid=self._sid,
rts_config=self._rts_config)
self._report.info('Validating and assigning resource manager')
if not self._rmgr._validate_resource_desc():
self._logger.error('Could not validate resource description')
raise ree.EnTKError('Could not validate resource description')
self._rmgr._populate()
self._rmgr.shared_data = self._shared_data
self._rmgr.outputs = self._outputs
self._report.ok('>>ok\n')
# --------------------------------------------------------------------------
#
@workflow.setter
def workflow(self, workflow):
self._prof.prof('assigning workflow', uid=self._uid)
for p in workflow:
if not isinstance(p, Pipeline):
self._logger.info('workflow type incorrect')
raise ree.TypeError(
expected_type=['Pipeline', 'set of Pipelines'],
actual_type=type(p))
p._validate()
# keep history
self._workflows.append(workflow)
# set current workflow
self._workflow = workflow
self._logger.info('Workflow assigned to Application Manager')
# --------------------------------------------------------------------------
#
@shared_data.setter
def shared_data(self, data):
if not isinstance(data, list):
data = [data]
for value in data:
if not isinstance(value, str):
raise ree.TypeError(expected_type=str, actual_type=type(value))
self._shared_data = data
if self._rmgr:
self._rmgr.shared_data = data
# --------------------------------------------------------------------------
#
@outputs.setter
def outputs(self, data):
if not isinstance(data, list):
data = [data]
for value in data:
if not isinstance(value, str):
raise ree.TypeError(expected_type=str, actual_type=type(value))
if self._rmgr:
self._rmgr.outputs = data
# --------------------------------------------------------------------------
#
# Public methods
#
def run(self):
'''
**Purpose**: Run the application manager. Once the workflow and resource
manager have been assigned. Invoking this method will start the setting
up the communication infrastructure, submitting a resource request and
then submission of all the tasks.
'''
ret = None
try:
self._prof.prof('amgr_start', uid=self._uid)
# Set None for variables local to each run
self._resubmit_failed = False
self._cur_attempt = 1
# Ensure that a workflow and a resource description have
# been defined
if not self._workflow:
self._logger.error('No workflow assignedcurrently, please \
check your script')
raise ree.MissingError(obj=self._uid,
missing_attribute='workflow')
if not self._rmgr:
self._logger.error('No resource manager assigned currently, \
please create and add a valid resource \
manager')
raise ree.MissingError(obj=self._uid,
missing_attribute='resource_manager')
self._prof.prof('amgr run started', uid=self._uid)
# ensure rabbitmq setup
self._setup_mqs()
# Submit resource request if no resource allocation done till now or
# resubmit a new one if the old one has completed
res_alloc_state = self._rmgr.get_resource_allocation_state()
if not res_alloc_state or \
res_alloc_state in self._rmgr.get_completed_states():
self._logger.info('Starting resource request submission')
self._prof.prof('rreq_init', uid=self._uid)
self._rmgr._submit_resource_request()
res_alloc_state = self._rmgr.get_resource_allocation_state()
if res_alloc_state in self._rmgr.get_completed_states():
raise ree.EnTKError(msg='Cannot proceed. Resource '
'ended in state %s' % res_alloc_state)
# Start all components and subcomponents
self._start_all_comps()
# Run workflow -- this call is blocking till all tasks of the
# workflow are executed or an error/exception is encountered
self._run_workflow()
self._logger.info('Workflow execution finished.')
if self._autoterminate:
self._logger.debug('Autoterminate set to %s.' %
self._autoterminate)
self.terminate()
except KeyboardInterrupt:
self._logger.exception('Execution interrupted by user (you '
'probably hit Ctrl+C), trying to cancel '
'enqueuer thread gracefully...')
self.terminate()
raise KeyboardInterrupt
except Exception:
self._logger.exception('Error in AppManager')
self.terminate()
raise
# return list of fetched output data, or None.
outputs = self.outputs
if outputs:
ret = outputs
return ret
# --------------------------------------------------------------------------
#
def terminate(self):
self._prof.prof('term_start', uid=self._uid)
# Terminate threads in following order: wfp, helper, synchronizer
if self._wfp:
self._logger.info('Terminating WFprocessor')
self._wfp.terminate_processor()
if self._task_manager:
self._logger.info('Terminating task manager process')
self._task_manager.terminate_manager()
self._task_manager.terminate_heartbeat()
if self._sync_thread:
self._logger.info('Terminating synchronizer thread')
self._terminate_sync.set()
self._sync_thread.join()
self._logger.info('Synchronizer thread terminated')
if self._write_workflow:
write_workflows(self.workflows, self._sid)
if self._rmgr:
self._rmgr._terminate_resource_request()
if os.environ.get('RADICAL_ENTK_PROFILE'):
write_session_description(self)
if self._rmq_cleanup:
self._cleanup_mqs()
self._report.info('All components terminated\n')
self._prof.prof('termination done', uid=self._uid)
# --------------------------------------------------------------------------
#
def resource_terminate(self):
self._logger.warning('DeprecationWarning: `resource_terminate()` is '
'deprecated, please use `terminate()`')
self.terminate()
# --------------------------------------------------------------------------
#
# Private methods
#
def _setup_mqs(self):
'''
**Purpose**: Setup RabbitMQ system on the client side. We instantiate
queue(s) 'pendingq-*' for communication between the enqueuer thread and
the task manager process. We instantiate queue(s) 'completedq-*' for
communication between the task manager and dequeuer thread. We
instantiate queue 'sync-to-master' for communication from
enqueuer/dequeuer/task_manager to the synchronizer thread. We
instantiate queue 'sync-ack' for communication from synchronizer thread
to enqueuer/dequeuer/task_manager.
Details: All queues are durable: Even if the RabbitMQ server goes down,
the queues are saved to disk and can be retrieved. This also means that
after an erroneous run the queues might still have unacknowledged
messages and will contain messages from that run. Hence, in every new
run, we first delete the queue and create a new one.
'''
try:
self._report.info('Setting up RabbitMQ system')
if self._mqs_setup:
self._report.ok('>>n/a\n')
return
self._report.ok('>>ok\n')
self._prof.prof('mqs_setup_start', uid=self._uid)
self._logger.debug('Setting up mq connection and channel')
mq_connection = pika.BlockingConnection(self._rmq_conn_params)
mq_channel = mq_connection.channel()
self._logger.debug('Connection and channel setup successful')
self._logger.debug('Setting up all exchanges and queues')
qs = [
'%s-tmgr-to-sync' % self._sid,
'%s-cb-to-sync' % self._sid,
'%s-sync-to-tmgr' % self._sid,
'%s-sync-to-cb' % self._sid
]
for i in range(1, self._num_pending_qs + 1):
queue_name = '%s-pendingq-%s' % (self._sid, i)
self._pending_queue.append(queue_name)
qs.append(queue_name)
for i in range(1, self._num_completed_qs + 1):
queue_name = '%s-completedq-%s' % (self._sid, i)
self._completed_queue.append(queue_name)
qs.append(queue_name)
# f = open('.%s.txt' % self._sid, 'w')
for q in qs:
# Durable Qs will not be lost if rabbitmq server crashes
mq_channel.queue_declare(queue=q)
# f.write(q + '\n')
# f.close()
self._mqs_setup = True
self._logger.debug('All exchanges and queues are setup')
self._prof.prof('mqs_setup_stop', uid=self._uid)
except Exception as ex:
self._logger.exception('Error setting RabbitMQ system: %s' % ex)
raise
# --------------------------------------------------------------------------
#
def _cleanup_mqs(self):
try:
self._prof.prof('mqs_cleanup_start', uid=self._uid)
mq_connection = pika.BlockingConnection(self._rmq_conn_params)
mq_channel = mq_connection.channel()
mq_channel.queue_delete(queue='%s-tmgr-to-sync' % self._sid)
mq_channel.queue_delete(queue='%s-cb-to-sync' % self._sid)
mq_channel.queue_delete(queue='%s-sync-to-tmgr' % self._sid)
mq_channel.queue_delete(queue='%s-sync-to-cb' % self._sid)
for i in range(1, self._num_pending_qs + 1):
queue_name = '%s-pendingq-%s' % (self._sid, i)
mq_channel.queue_delete(queue=queue_name)
for i in range(1, self._num_completed_qs + 1):
queue_name = '%s-completedq-%s' % (self._sid, i)
mq_channel.queue_delete(queue=queue_name)
self._prof.prof('mqs_cleanup_stop', uid=self._uid)
self._mqs_setup = False
except Exception:
self._logger.exception('Message queues not deleted, error')
raise
# --------------------------------------------------------------------------
#
def _start_all_comps(self):
if self._wfp:
# This condition is called when there are multiple workflows
# submitted for execution. Amgr.run() was probably called twice.
# If a WFP exists, we use the same one but with the new workflow.
# Since WFP (and its threads) and the Amgr share memory, we have
# to terminate WFP's threads, assign the new workflow and then
# start the threads again.
self._wfp.terminate_processor()
self._wfp._workflow = self._workflow
self._wfp.start_processor()
return
# Create WFProcessor and initialize workflow its contents with
# uids
self._prof.prof('wfp_create_start', uid=self._uid)
self._wfp = WFprocessor(sid=self._sid,
workflow=self._workflow,
pending_queue=self._pending_queue,
completed_queue=self._completed_queue,
resubmit_failed=self._resubmit_failed,
rmq_conn_params=self._rmq_conn_params)
self._prof.prof('wfp_create_stop', uid=self._uid)
# Start synchronizer thread AM OK
if not self._sync_thread:
self._logger.info('Starting synchronizer thread')
self._sync_thread = mt.Thread(target=self._synchronizer,
name='synchronizer-thread')
self._prof.prof('sync_thread_create', uid=self._uid)
self._sync_thread.start()
# Start WFprocessor
self._logger.info('Starting WFProcessor')
self._wfp.start_processor()
self._report.ok('All components created\n')
# Create tmgr object only if it does not already exist
if self._rts == 'radical.pilot':
from radical.entk.execman.rp import TaskManager
elif self._rts == 'mock':
from radical.entk.execman.mock import TaskManager
if not self._task_manager:
self._logger.info('Starting task manager')
self._prof.prof('tmgr_create_start', uid=self._uid)
self._task_manager = TaskManager(
sid=self._sid,
pending_queue=self._pending_queue,
completed_queue=self._completed_queue,
rmgr=self._rmgr,
rmq_conn_params=self._rmq_conn_params)
self._task_manager.start_manager()
self._task_manager.start_heartbeat()
self._prof.prof('tmgr_create_stop', uid=self._uid)
# --------------------------------------------------------------------------
#
def _run_workflow(self):
active_pipe_count = len(self._workflow)
finished_pipe_uids = list()
# We wait till all pipelines of the workflow are marked
# complete
state = self._rmgr.get_resource_allocation_state()
final = self._rmgr.get_completed_states()
incomplete = self._wfp.workflow_incomplete()
while active_pipe_count and \
incomplete and \
state not in final:
state = self._rmgr.get_resource_allocation_state()
for pipe in self._workflow:
with pipe.lock:
if pipe.completed and \
pipe.uid not in finished_pipe_uids:
finished_pipe_uids.append(pipe.uid)
active_pipe_count -= 1
self._logger.info('Pipe %s completed' % pipe.uid)
self._logger.info('Active pipes %s' %
active_pipe_count)
if not self._sync_thread.is_alive() and \
self._cur_attempt <= self._reattempts:
self._sync_thread = mt.Thread(target=self._synchronizer,
name='synchronizer-thread')
self._sync_thread.start()
self._cur_attempt += 1
self._prof.prof('sync_thread_restart', uid=self._uid)
self._logger.info('Restarting synchronizer thread')
if not self._wfp.check_processor() and \
self._cur_attempt <= self._reattempts:
# If WFP dies, both child threads are also cleaned out.
# We simply recreate the wfp object with a copy of the
# workflow in the appmanager and start the processor.
self._prof.prof('wfp_recreate', uid=self._uid)
self._wfp = WFprocessor(sid=self._sid,
workflow=self._workflow,
pending_queue=self._pending_queue,
completed_queue=self._completed_queue,
resubmit_failed=self._resubmit_failed,
rmq_conn_params=self._rmq_conn_params)
self._logger.info('Restarting WFProcessor')
self._wfp.start_processor()
self._cur_attempt += 1
if not self._task_manager.check_heartbeat() and \
self._cur_attempt <= self._reattempts:
# If the tmgr process or heartbeat dies, we simply start a
# new process using the start_manager method. We do not
# need to create a new instance of the TaskManager object
# itself. We stop and start a new instance of the
# heartbeat thread as well.
self._prof.prof('restart_tmgr', uid=self._uid)
self._logger.info('Terminating heartbeat thread')
self._task_manager.terminate_heartbeat()
self._logger.info('Terminating tmgr process')
self._task_manager.terminate_manager()
self._logger.info('Restarting task manager process')
self._task_manager.start_manager()
self._logger.info('Restarting heartbeat thread')
self._task_manager.start_heartbeat()
self._cur_attempt += 1
# --------------------------------------------------------------------------
#
def _get_message_to_sync(self, mq_channel, qname):
'''
Reads a message from the queue, and exchange the message to where it
was published by `update_task`
'''
# --------------------------------------------------------------
# Messages between tmgr Main thread and synchronizer -- only
# Task objects
method_frame, props, body = mq_channel.basic_get(queue=qname)
tmp = qname.split("-")
q_sid = ''.join(tmp[:-3])
q_from = tmp[-3]
q_to = tmp[-1]
return_queue_name = f"{q_sid}-{q_to}-to-{q_from}"
# The message received is a JSON object with the following
# structure:
# msg = {
# 'type': 'Pipeline'/'Stage'/'Task',
# 'object': json/dict
# }
if body:
msg = json.loads(body)
uid = msg['object']['uid']
state = msg['object']['state']
self._prof.prof('sync_recv_obj_state_%s' % state, uid=uid)
self._logger.debug('recv %s in state %s (sync)' % (uid, state))
if msg['type'] == 'Task':
self._update_task(msg, return_queue_name, props.correlation_id,
mq_channel, method_frame)
# --------------------------------------------------------------------------
#
def _update_task(self, msg, reply_to, corr_id, mq_channel, method_frame):
# pylint: disable=W0612,W0613
completed_task = Task()
completed_task.from_dict(msg['object'])
self._logger.info('Received %s with state %s' %
(completed_task.uid, completed_task.state))
# found_task = False
# Traverse the entire workflow to find the correct task
for pipe in self._workflow:
with pipe.lock:
if pipe.completed or \
pipe.uid != completed_task.parent_pipeline['uid']:
continue
for stage in pipe.stages:
if stage.uid != completed_task.parent_stage['uid']:
continue
for task in stage.tasks:
if completed_task.uid != task.uid or \
completed_task.state == task.state:
continue
self._logger.debug(
('Found task %s in state (%s)'
' changing to %s ==') %
(task.uid, task.state, completed_task.state))
if task.state in [states.DONE, states.FAILED]:
self._logger.debug(
('No change on task state %s '
'in state %s') % (task.uid, task.state))
break
task.state = str(completed_task.state)
self._logger.debug('Found task %s in state %s' %
(task.uid, task.state))
if completed_task.path:
task.path = str(completed_task.path)
# mq_channel.basic_publish(
# exchange='',
# routing_key=reply_to,
# properties=pika.BasicProperties(
# correlation_id=corr_id),
# body='%s-ack' % task.uid)
state = msg['object']['state']
self._prof.prof('pub_ack_state_%s' % state,
uid=msg['object']['uid'])
mq_channel.basic_ack(
delivery_tag=method_frame.delivery_tag)
self._report.ok('Update: ')
self._report.info('%s state: %s\n' %
(task.luid, task.state))
# found_task = True
break
# if not found_task:
#
# # If there was a Stage update, but the Stage was
# # not found in any of the Pipelines. This
# # means that this was a Stage that was added
# # during runtime and the AppManager does not
# # know about it. The current solution is going
# # to be: add it to the workflow object in the
# # AppManager via the synchronizer.
#
# self._logger.info('Adding new task %s to \
# parent stage: %s'
# % (completed_task.uid,
# stage.uid))
#
# self._prof.prof('adap_add_task_start',
# uid=completed_task.uid)
# stage.add_tasks(completed_task)
# self._prof.prof('adap_add_task_stop',
# uid=completed_task.uid)
#
# mq_channel.basic_publish(exchange='',
# routing_key=reply_to,
# properties=pika.BasicProperties(
# correlation_id=corr_id),
# body='%s-ack' % completed_task.uid)
#
# self._prof.prof('pub_ack_state_%s' %
# msg['object']['state'],
# uid=msg['object']['uid'])
#
# mq_channel.basic_ack(
# delivery_tag=method_frame.delivery_tag)
# self._report.ok('Update: ')
# self._report.info('%s state: %s\n' %
# (completed_task.luid, completed_task.state))
# --------------------------------------------------------------------------
#
def _synchronizer(self):
try:
self._synchronizer_work()
except KeyboardInterrupt:
self._logger.exception('Execution interrupted by user (you \
probably hit Ctrl+C), trying to terminate \
synchronizer thread gracefully...')
raise
except Exception:
self._logger.exception('Unknown error in synchronizer: %s. \
Terminating thread')
raise
# --------------------------------------------------------------------------
#
def _synchronizer_work(self):
'''
**Purpose**: Thread in the master process to keep the workflow data
structure in appmanager up to date. We receive only tasks
objects from the task manager.
Details: Important to note that acknowledgements of the type
`channel.basic_ack()` is an acknowledgement to the server
that the msg was received. This is not to be confused with
the Ack sent to the task_manager through the sync-ack
queue.
'''
self._prof.prof('sync_thread_start', uid=self._uid)
self._logger.info('synchronizer thread started')
mq_connection = pika.BlockingConnection(self._rmq_conn_params)
mq_channel = mq_connection.channel()
last = time.time()
qname_t2s = '%s-tmgr-to-sync' % self._sid
qname_c2s = '%s-cb-to-sync' % self._sid
while not self._terminate_sync.is_set():
# wrapper to call `_update_task()`
self._get_message_to_sync(mq_channel, qname_t2s)
self._get_message_to_sync(mq_channel, qname_c2s)
# Appease pika cos it thinks the connection is dead
now = time.time()
if now - last >= self._rmq_ping_interval:
mq_connection.process_data_events()
last = now
self._prof.prof('sync_thread_stop', uid=self._uid)
def run(self):
"""
**Purpose**: Run the application manager. Once the workflow and resource manager have been assigned. Invoking this
method will start the setting up the communication infrastructure, submitting a resource request and then
submission of all the tasks.
"""
try:
# Set None objects local to each run
self._wfp = None
self._sync_thread = None
self._terminate_sync = Event()
self._resubmit_failed = False
self._cur_attempt = 1
if not self._workflow:
self._logger.error(
'No workflow assigned currently, please check your script')
raise MissingError(obj=self._uid, missing_attribute='workflow')
if not self._resource_manager:
self._logger.error(
'No resource manager assigned currently, please create and add a valid resource manager'
)
raise MissingError(obj=self._uid,
missing_attribute='resource_manager')
self._prof.prof('amgr run started', uid=self._uid)
# Setup rabbitmq stuff
if not self._mqs_setup:
self._report.info('Setting up RabbitMQ system')
setup = self._setup_mqs()
if not setup:
self._logger.error('RabbitMQ system not available')
raise EnTKError("RabbitMQ setup failed")
self._mqs_setup = True
self._report.ok('>>ok\n')
# Create WFProcessor object
self._prof.prof('creating wfp obj', uid=self._uid)
self._wfp = WFprocessor(sid=self._sid,
workflow=self._workflow,
pending_queue=self._pending_queue,
completed_queue=self._completed_queue,
mq_hostname=self._mq_hostname,
port=self._port,
resubmit_failed=self._resubmit_failed)
self._wfp._initialize_workflow()
self._workflow = self._wfp.workflow
# Submit resource request if not resource allocation done till now or
# resubmit a new one if the old one has completed
if self._resource_manager:
res_alloc_state = self._resource_manager.get_resource_allocation_state(
)
if (not res_alloc_state) or (
res_alloc_state
in self._resource_manager.get_completed_states()):
self._logger.info('Starting resource request submission')
self._prof.prof('init rreq submission', uid=self._uid)
self._resource_manager._submit_resource_request()
else:
self._logger.error(
'Cannot run without resource manager, please create and assign a resource manager'
)
raise EnTKError(text='Missing resource manager')
# Start synchronizer thread
if not self._sync_thread:
self._logger.info('Starting synchronizer thread')
self._sync_thread = Thread(target=self._synchronizer,
name='synchronizer-thread')
self._prof.prof('starting synchronizer thread', uid=self._uid)
self._sync_thread.start()
# Start WFprocessor
self._logger.info('Starting WFProcessor process from AppManager')
self._wfp.start_processor()
self._report.ok('All components created\n')
# Create tmgr object only if it does not already exist
if self._rts == 'radical.pilot':
from radical.entk.execman.rp import TaskManager
elif self._rts == 'mock':
from radical.entk.execman.mock import TaskManager
if not self._task_manager:
self._prof.prof('creating tmgr obj', uid=self._uid)
self._task_manager = TaskManager(
sid=self._sid,
pending_queue=self._pending_queue,
completed_queue=self._completed_queue,
mq_hostname=self._mq_hostname,
rmgr=self._resource_manager,
port=self._port)
self._logger.info(
'Starting task manager process from AppManager')
self._task_manager.start_manager()
self._task_manager.start_heartbeat()
active_pipe_count = len(self._workflow)
finished_pipe_uids = []
# We wait till all pipelines of the workflow are marked
# complete
while ((active_pipe_count > 0)
and (self._wfp.workflow_incomplete())
and (self._resource_manager.get_resource_allocation_state()
not in self._resource_manager.get_completed_states())):
if active_pipe_count > 0:
for pipe in self._workflow:
with pipe.lock:
if (pipe.completed) and (
pipe.uid not in finished_pipe_uids):
self._logger.info('Pipe %s completed' %
pipe.uid)
finished_pipe_uids.append(pipe.uid)
active_pipe_count -= 1
self._logger.info('Active pipes: %s' %
active_pipe_count)
if (not self._sync_thread.is_alive()) and (self._cur_attempt <=
self._reattempts):
self._sync_thread = Thread(target=self._synchronizer,
name='synchronizer-thread')
self._logger.info('Restarting synchronizer thread')
self._prof.prof('restarting synchronizer', uid=self._uid)
self._sync_thread.start()
self._cur_attempt += 1
if (not self._wfp.check_processor()) and (self._cur_attempt <=
self._reattempts):
"""
If WFP dies, both child threads are also cleaned out.
We simply recreate the wfp object with a copy of the workflow
in the appmanager and start the processor.
"""
self._prof.prof('recreating wfp obj', uid=self._uid)
self._wfp = WFProcessor(
sid=self._sid,
workflow=self._workflow,
pending_queue=self._pending_queue,
completed_queue=self._completed_queue,
mq_hostname=self._mq_hostname,
port=self._port,
resubmit_failed=self._resubmit_failed)
self._logger.info(
'Restarting WFProcessor process from AppManager')
self._wfp.start_processor()
self._cur_attempt += 1
if (not self._task_manager.check_heartbeat()) and (
self._cur_attempt <= self._reattempts):
"""
If the tmgr process or heartbeat dies, we simply start a
new process using the start_manager method. We do not
need to create a new instance of the TaskManager object
itself. We stop and start a new instance of the
heartbeat thread as well.
"""
self._prof.prof('restarting tmgr process and heartbeat',
uid=self._uid)
self._logger.info('Terminating heartbeat thread')
self._task_manager.terminate_heartbeat()
self._logger.info('Terminating tmgr process')
self._task_manager.terminate_manager()
self._logger.info('Restarting task manager process')
self._task_manager.start_manager()
self._logger.info('Restarting heartbeat thread')
self._task_manager.start_heartbeat()
self._cur_attempt += 1
self._prof.prof('start termination', uid=self._uid)
# Terminate threads in following order: wfp, helper, synchronizer
self._logger.info('Terminating WFprocessor')
self._wfp.terminate_processor()
self._logger.info('Terminating synchronizer thread')
self._terminate_sync.set()
self._sync_thread.join()
self._logger.info('Synchronizer thread terminated')
if self._autoterminate:
self.resource_terminate()
if self._write_workflow:
write_workflow(self._workflow, self._sid)
self._prof.prof('termination done', uid=self._uid)
except KeyboardInterrupt:
self._prof.prof('start termination', uid=self._uid)
self._logger.error(
'Execution interrupted by user (you probably hit Ctrl+C), ' +
'trying to cancel enqueuer thread gracefully...')
# Terminate threads in following order: wfp, helper, synchronizer
if self._wfp:
self._logger.info('Terminating WFprocessor')
self._wfp.terminate_processor()
if self._task_manager:
self._logger.info('Terminating task manager process')
self._task_manager.terminate_manager()
self._task_manager.terminate_heartbeat()
if self._sync_thread:
self._logger.info('Terminating synchronizer thread')
self._terminate_sync.set()
self._sync_thread.join()
self._logger.info('Synchronizer thread terminated')
if self._resource_manager:
self._resource_manager._terminate_resource_request()
self._prof.prof('termination done', uid=self._uid)
raise KeyboardInterrupt
except Exception, ex:
self._prof.prof('start termination', uid=self._uid)
self._logger.exception('Error in AppManager: %s' % ex)
# Terminate threads in following order: wfp, helper, synchronizer
if self._wfp:
self._logger.info('Terminating WFprocessor')
self._wfp.terminate_processor()
if self._task_manager:
self._logger.info('Terminating task manager process')
self._task_manager.terminate_manager()
self._task_manager.terminate_heartbeat()
if self._sync_thread:
self._logger.info('Terminating synchronizer thread')
self._terminate_sync.set()
self._sync_thread.join()
self._logger.info('Synchronizer thread terminated')
if self._resource_manager:
self._resource_manager._terminate_resource_request()
self._prof.prof('termination done', uid=self._uid)
raise
class AppManager(object):
"""
An application manager takes the responsibility of setting up the communication infrastructure, instantiates the
ResourceManager, TaskManager, WFProcessor objects and all their threads and processes. This is the Master object
running in the main process and is designed to recover from errors from all other objects, threads and processes.
:Arguments:
:config_path: Url to config path to be read for AppManager
:hostname: host rabbitmq server is running
:port: port at which rabbitmq can be accessed
:reattempts: number of attempts to re-invoke any failed EnTK components
:resubmit_failed: resubmit failed tasks (True/False)
:autoterminate: terminate resource reservation upon execution of all tasks of first workflow (True/False)
:write_workflow: write workflow and mapping to rts entities to a file (post-termination)
:rts: Specify RTS to use. Current options: 'mock', 'radical.pilot' (default if unspecified)
:rmq_cleanup: Cleanup all queues created in RabbitMQ server for current execution (default is True)
:rts_config: Configuration for the RTS, accepts {"sandbox_cleanup": True/False,"db_cleanup": True/False} when RTS is RP
:name: Name of the Application. It should be unique between executions. (default is randomly assigned)
"""
def __init__(self,
config_path=None,
hostname=None,
port=None,
reattempts=None,
resubmit_failed=None,
autoterminate=None,
write_workflow=None,
rts=None,
rmq_cleanup=None,
rts_config=None,
name=None):
# Create a session for each EnTK script execution
if name:
self._name = name
self._sid = name
else:
self._name = str()
self._sid = ru.generate_id('re.session', ru.ID_PRIVATE)
self._read_config(config_path, hostname, port, reattempts,
resubmit_failed, autoterminate, write_workflow, rts,
rmq_cleanup, rts_config)
# Create an uid + logger + profiles for AppManager, under the sid
# namespace
path = os.getcwd() + '/' + self._sid
self._uid = ru.generate_id('appmanager.%(item_counter)04d',
ru.ID_CUSTOM,
namespace=self._sid)
self._logger = ru.Logger('radical.entk.%s' % self._uid,
path=path,
targets=['2', '.'])
self._prof = ru.Profiler(name='radical.entk.%s' % self._uid, path=path)
self._report = ru.Reporter(name='radical.entk.%s' % self._uid)
self._report.info('EnTK session: %s\n' % self._sid)
self._prof.prof('create amgr obj', uid=self._uid)
self._report.info('Creating AppManager')
self._resource_manager = None
# RabbitMQ Queues
self._pending_queue = list()
self._completed_queue = list()
# Global parameters to have default values
self._mqs_setup = False
self._resource_desc = None
self._task_manager = None
self._workflow = None
self._cur_attempt = 1
self._shared_data = list()
self._rmq_ping_interval = os.getenv('RMQ_PING_INTERVAL', 10)
self._logger.info('Application Manager initialized')
self._prof.prof('amgr obj created', uid=self._uid)
self._report.ok('>>ok\n')
def _read_config(self, config_path, hostname, port, reattempts,
resubmit_failed, autoterminate, write_workflow, rts,
rmq_cleanup, rts_config):
if not config_path:
config_path = os.path.dirname(os.path.abspath(__file__))
config = ru.read_json(os.path.join(config_path, 'config.json'))
self._mq_hostname = hostname if hostname else str(config['hostname'])
self._port = port if port else config['port']
self._reattempts = reattempts if reattempts else config['reattempts']
self._resubmit_failed = resubmit_failed if resubmit_failed is not None else config[
'resubmit_failed']
self._autoterminate = autoterminate if autoterminate is not None else config[
'autoterminate']
self._write_workflow = write_workflow if write_workflow is not None else config[
'write_workflow']
self._rts = rts if rts in ['radical.pilot', 'mock'] else str(
config['rts'])
self._rmq_cleanup = rmq_cleanup if rmq_cleanup is not None else config[
'rmq_cleanup']
self._rts_config = rts_config if rts_config is not None else config[
'rts_config']
self._num_pending_qs = config['pending_qs']
self._num_completed_qs = config['completed_qs']
# ------------------------------------------------------------------------------------------------------------------
# Getter functions
# ------------------------------------------------------------------------------------------------------------------
@property
def name(self):
"""
Name for the application manager. Allows the user to setup the name of
the application manager, as well as, its session ID. This name should be
unique between different EnTK executions, otherwise it will produce an
error.
:getter: Returns the name of the application manager
:setter: Assigns the name of the application manager
:type: String
"""
return self._name
@property
def sid(self):
"""
Get the session ID of the current EnTK execution
:getter: Returns the session ID of the EnTK execution
:type: String
"""
return self._sid
@property
def resource_desc(self):
"""
:getter: Returns the resource description
:setter: Assigns a resource description
"""
return self._resource_desc
@property
def workflow(self):
"""
:getter: Return the workflow assigned for execution
:setter: Assign workflow to be executed
"""
return self._workflow
@property
def shared_data(self):
"""
:getter: Return list of filenames that are shared between multiple tasks of the application
:setter: Assign a list of names of files that need to be staged to the remote machine
"""
return self._shared_data
# ------------------------------------------------------------------------------------------------------------------
# Setter functions
# ------------------------------------------------------------------------------------------------------------------
@name.setter
def name(self, value):
if not isinstance(value, str):
raise TypeError(expected_type=str, actual_type=type(value))
else:
self._name = value
@resource_desc.setter
def resource_desc(self, value):
if self._rts == 'radical.pilot':
from radical.entk.execman.rp import ResourceManager
self._resource_manager = ResourceManager(
resource_desc=value,
sid=self._sid,
rts_config=self._rts_config)
elif self._rts == 'mock':
from radical.entk.execman.mock import ResourceManager
self._resource_manager = ResourceManager(resource_desc=value,
sid=self._sid)
self._report.info('Validating and assigning resource manager')
if self._resource_manager._validate_resource_desc():
self._resource_manager._populate()
self._resource_manager.shared_data = self._shared_data
else:
self._logger.error('Could not validate resource description')
raise
self._report.ok('>>ok\n')
@workflow.setter
def workflow(self, workflow):
self._prof.prof('assigning workflow', uid=self._uid)
for p in workflow:
if not isinstance(p, Pipeline):
self._logger.info('workflow type incorrect')
raise TypeError(expected_type=['Pipeline', 'set of Pipelines'],
actual_type=type(p))
p._validate()
self._workflow = workflow
self._logger.info('Workflow assigned to Application Manager')
@shared_data.setter
def shared_data(self, data):
if not isinstance(data, list):
data = [data]
for val in data:
if not isinstance(val, str):
raise TypeError(expected_type=str, actual_type=type(val))
if self._resource_manager:
self._resource_manager.shared_data = data
# ------------------------------------------------------------------------------------------------------------------
# Public methods
# ------------------------------------------------------------------------------------------------------------------
def run(self):
"""
**Purpose**: Run the application manager. Once the workflow and resource manager have been assigned. Invoking this
method will start the setting up the communication infrastructure, submitting a resource request and then
submission of all the tasks.
"""
try:
# Set None objects local to each run
self._wfp = None
self._sync_thread = None
self._terminate_sync = Event()
self._resubmit_failed = False
self._cur_attempt = 1
if not self._workflow:
self._logger.error(
'No workflow assigned currently, please check your script')
raise MissingError(obj=self._uid, missing_attribute='workflow')
if not self._resource_manager:
self._logger.error(
'No resource manager assigned currently, please create and add a valid resource manager'
)
raise MissingError(obj=self._uid,
missing_attribute='resource_manager')
self._prof.prof('amgr run started', uid=self._uid)
# Setup rabbitmq stuff
if not self._mqs_setup:
self._report.info('Setting up RabbitMQ system')
setup = self._setup_mqs()
if not setup:
self._logger.error('RabbitMQ system not available')
raise EnTKError("RabbitMQ setup failed")
self._mqs_setup = True
self._report.ok('>>ok\n')
# Create WFProcessor object
self._prof.prof('creating wfp obj', uid=self._uid)
self._wfp = WFprocessor(sid=self._sid,
workflow=self._workflow,
pending_queue=self._pending_queue,
completed_queue=self._completed_queue,
mq_hostname=self._mq_hostname,
port=self._port,
resubmit_failed=self._resubmit_failed)
self._wfp._initialize_workflow()
self._workflow = self._wfp.workflow
# Submit resource request if not resource allocation done till now or
# resubmit a new one if the old one has completed
if self._resource_manager:
res_alloc_state = self._resource_manager.get_resource_allocation_state(
)
if (not res_alloc_state) or (
res_alloc_state
in self._resource_manager.get_completed_states()):
self._logger.info('Starting resource request submission')
self._prof.prof('init rreq submission', uid=self._uid)
self._resource_manager._submit_resource_request()
else:
self._logger.error(
'Cannot run without resource manager, please create and assign a resource manager'
)
raise EnTKError(text='Missing resource manager')
# Start synchronizer thread
if not self._sync_thread:
self._logger.info('Starting synchronizer thread')
self._sync_thread = Thread(target=self._synchronizer,
name='synchronizer-thread')
self._prof.prof('starting synchronizer thread', uid=self._uid)
self._sync_thread.start()
# Start WFprocessor
self._logger.info('Starting WFProcessor process from AppManager')
self._wfp.start_processor()
self._report.ok('All components created\n')
# Create tmgr object only if it does not already exist
if self._rts == 'radical.pilot':
from radical.entk.execman.rp import TaskManager
elif self._rts == 'mock':
from radical.entk.execman.mock import TaskManager
if not self._task_manager:
self._prof.prof('creating tmgr obj', uid=self._uid)
self._task_manager = TaskManager(
sid=self._sid,
pending_queue=self._pending_queue,
completed_queue=self._completed_queue,
mq_hostname=self._mq_hostname,
rmgr=self._resource_manager,
port=self._port)
self._logger.info(
'Starting task manager process from AppManager')
self._task_manager.start_manager()
self._task_manager.start_heartbeat()
active_pipe_count = len(self._workflow)
finished_pipe_uids = []
# We wait till all pipelines of the workflow are marked
# complete
while ((active_pipe_count > 0)
and (self._wfp.workflow_incomplete())
and (self._resource_manager.get_resource_allocation_state()
not in self._resource_manager.get_completed_states())):
if active_pipe_count > 0:
for pipe in self._workflow:
with pipe.lock:
if (pipe.completed) and (
pipe.uid not in finished_pipe_uids):
self._logger.info('Pipe %s completed' %
pipe.uid)
finished_pipe_uids.append(pipe.uid)
active_pipe_count -= 1
self._logger.info('Active pipes: %s' %
active_pipe_count)
if (not self._sync_thread.is_alive()) and (self._cur_attempt <=
self._reattempts):
self._sync_thread = Thread(target=self._synchronizer,
name='synchronizer-thread')
self._logger.info('Restarting synchronizer thread')
self._prof.prof('restarting synchronizer', uid=self._uid)
self._sync_thread.start()
self._cur_attempt += 1
if (not self._wfp.check_processor()) and (self._cur_attempt <=
self._reattempts):
"""
If WFP dies, both child threads are also cleaned out.
We simply recreate the wfp object with a copy of the workflow
in the appmanager and start the processor.
"""
self._prof.prof('recreating wfp obj', uid=self._uid)
self._wfp = WFProcessor(
sid=self._sid,
workflow=self._workflow,
pending_queue=self._pending_queue,
completed_queue=self._completed_queue,
mq_hostname=self._mq_hostname,
port=self._port,
resubmit_failed=self._resubmit_failed)
self._logger.info(
'Restarting WFProcessor process from AppManager')
self._wfp.start_processor()
self._cur_attempt += 1
if (not self._task_manager.check_heartbeat()) and (
self._cur_attempt <= self._reattempts):
"""
If the tmgr process or heartbeat dies, we simply start a
new process using the start_manager method. We do not
need to create a new instance of the TaskManager object
itself. We stop and start a new instance of the
heartbeat thread as well.
"""
self._prof.prof('restarting tmgr process and heartbeat',
uid=self._uid)
self._logger.info('Terminating heartbeat thread')
self._task_manager.terminate_heartbeat()
self._logger.info('Terminating tmgr process')
self._task_manager.terminate_manager()
self._logger.info('Restarting task manager process')
self._task_manager.start_manager()
self._logger.info('Restarting heartbeat thread')
self._task_manager.start_heartbeat()
self._cur_attempt += 1
self._prof.prof('start termination', uid=self._uid)
# Terminate threads in following order: wfp, helper, synchronizer
self._logger.info('Terminating WFprocessor')
self._wfp.terminate_processor()
self._logger.info('Terminating synchronizer thread')
self._terminate_sync.set()
self._sync_thread.join()
self._logger.info('Synchronizer thread terminated')
if self._autoterminate:
self.resource_terminate()
if self._write_workflow:
write_workflow(self._workflow, self._sid)
self._prof.prof('termination done', uid=self._uid)
except KeyboardInterrupt:
self._prof.prof('start termination', uid=self._uid)
self._logger.error(
'Execution interrupted by user (you probably hit Ctrl+C), ' +
'trying to cancel enqueuer thread gracefully...')
# Terminate threads in following order: wfp, helper, synchronizer
if self._wfp:
self._logger.info('Terminating WFprocessor')
self._wfp.terminate_processor()
if self._task_manager:
self._logger.info('Terminating task manager process')
self._task_manager.terminate_manager()
self._task_manager.terminate_heartbeat()
if self._sync_thread:
self._logger.info('Terminating synchronizer thread')
self._terminate_sync.set()
self._sync_thread.join()
self._logger.info('Synchronizer thread terminated')
if self._resource_manager:
self._resource_manager._terminate_resource_request()
self._prof.prof('termination done', uid=self._uid)
raise KeyboardInterrupt
except Exception, ex:
self._prof.prof('start termination', uid=self._uid)
self._logger.exception('Error in AppManager: %s' % ex)
# Terminate threads in following order: wfp, helper, synchronizer
if self._wfp:
self._logger.info('Terminating WFprocessor')
self._wfp.terminate_processor()
if self._task_manager:
self._logger.info('Terminating task manager process')
self._task_manager.terminate_manager()
self._task_manager.terminate_heartbeat()
if self._sync_thread:
self._logger.info('Terminating synchronizer thread')
self._terminate_sync.set()
self._sync_thread.join()
self._logger.info('Synchronizer thread terminated')
if self._resource_manager:
self._resource_manager._terminate_resource_request()
self._prof.prof('termination done', uid=self._uid)
raise
def test_write_session_description():
hostname = os.environ.get('RMQ_HOSTNAME', 'localhost')
port = int(os.environ.get('RMQ_PORT', 5672))
amgr = AppManager(hostname=hostname, port=port)
amgr.resource_desc = {
'resource': 'xsede.stampede',
'walltime': 60,
'cpus': 128,
'gpus': 64,
'project': 'xyz',
'queue': 'high'
}
workflow = [generate_pipeline(1), generate_pipeline(2)]
amgr.workflow = workflow
amgr._wfp = WFprocessor(sid=amgr._sid,
workflow=amgr._workflow,
pending_queue=amgr._pending_queue,
completed_queue=amgr._completed_queue,
mq_hostname=amgr._mq_hostname,
port=amgr._port,
resubmit_failed=amgr._resubmit_failed)
amgr._wfp._initialize_workflow()
amgr._workflow = amgr._wfp.workflow
amgr._task_manager = TaskManager(sid=amgr._sid,
pending_queue=amgr._pending_queue,
completed_queue=amgr._completed_queue,
mq_hostname=amgr._mq_hostname,
rmgr=amgr._resource_manager,
port=amgr._port)
# os.mkdir(amgr._sid)
write_session_description(amgr)
desc = ru.read_json('%s/radical.entk.%s.json' % (amgr._sid, amgr._sid))
assert desc == {
'config': {},
'entities': {
'appmanager': {
'event_model': {},
'state_model': None,
'state_values': None
},
'pipeline': {
'event_model': {},
'state_model': {
'CANCELED': 9,
'DESCRIBED': 1,
'DONE': 9,
'FAILED': 9,
'SCHEDULING': 2
},
'state_values': {
'1': 'DESCRIBED',
'2': 'SCHEDULING',
'9': ['DONE', 'CANCELED', 'FAILED']
}
},
'stage': {
'event_model': {},
'state_model': {
'CANCELED': 9,
'DESCRIBED': 1,
'DONE': 9,
'FAILED': 9,
'SCHEDULED': 3,
'SCHEDULING': 2
},
'state_values': {
'1': 'DESCRIBED',
'2': 'SCHEDULING',
'3': 'SCHEDULED',
'9': ['FAILED', 'CANCELED', 'DONE']
}
},
'task': {
'event_model': {},
'state_model': {
'CANCELED': 9,
'DEQUEUED': 8,
'DEQUEUEING': 7,
'DESCRIBED': 1,
'DONE': 9,
'EXECUTED': 6,
'FAILED': 9,
'SCHEDULED': 3,
'SCHEDULING': 2,
'SUBMITTED': 5,
'SUBMITTING': 4
},
'state_values': {
'1': 'DESCRIBED',
'2': 'SCHEDULING',
'3': 'SCHEDULED',
'4': 'SUBMITTING',
'5': 'SUBMITTED',
'6': 'EXECUTED',
'7': 'DEQUEUEING',
'8': 'DEQUEUED',
'9': ['DONE', 'CANCELED', 'FAILED']
}
}
},
'tree': {
'appmanager.0000': {
'cfg': {},
'children': [
'wfprocessor.0000', 'resource_manager.0000',
'task_manager.0000', 'pipeline.0000', 'pipeline.0001'
],
'etype':
'appmanager',
'has': [
'pipeline', 'wfprocessor', 'resource_manager',
'task_manager'
],
'uid':
'appmanager.0000'
},
'pipeline.0000': {
'cfg': {},
'children': ['stage.0000', 'stage.0001'],
'etype': 'pipeline',
'has': ['stage'],
'uid': 'pipeline.0000'
},
'pipeline.0001': {
'cfg': {},
'children': ['stage.0002', 'stage.0003'],
'etype': 'pipeline',
'has': ['stage'],
'uid': 'pipeline.0001'
},
'resource_manager.0000': {
'cfg': {},
'children': [],
'etype': 'resource_manager',
'has': [],
'uid': 'resource_manager.0000'
},
'stage.0000': {
'cfg': {},
'children': ['task.0000'],
'etype': 'stage',
'has': ['task'],
'uid': 'stage.0000'
},
'stage.0001': {
'cfg': {},
'children': [
'task.0001', 'task.0002', 'task.0003', 'task.0004',
'task.0005', 'task.0006', 'task.0007', 'task.0008',
'task.0009', 'task.0010'
],
'etype':
'stage',
'has': ['task'],
'uid':
'stage.0001'
},
'stage.0002': {
'cfg': {},
'children': ['task.0011'],
'etype': 'stage',
'has': ['task'],
'uid': 'stage.0002'
},
'stage.0003': {
'cfg': {},
'children': [
'task.0012', 'task.0013', 'task.0014', 'task.0015',
'task.0016', 'task.0017', 'task.0018', 'task.0019',
'task.0020', 'task.0021'
],
'etype':
'stage',
'has': ['task'],
'uid':
'stage.0003'
},
'task.0000': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0000'
},
'task.0001': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0001'
},
'task.0002': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0002'
},
'task.0003': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0003'
},
'task.0004': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0004'
},
'task.0005': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0005'
},
'task.0006': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0006'
},
'task.0007': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0007'
},
'task.0008': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0008'
},
'task.0009': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0009'
},
'task.0010': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0010'
},
'task.0011': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0011'
},
'task.0012': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0012'
},
'task.0013': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0013'
},
'task.0014': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0014'
},
'task.0015': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0015'
},
'task.0016': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0016'
},
'task.0017': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0017'
},
'task.0018': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0018'
},
'task.0019': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0019'
},
'task.0020': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0020'
},
'task.0021': {
'cfg': {},
'children': [],
'etype': 'task',
'has': [],
'uid': 'task.0021'
},
'task_manager.0000': {
'cfg': {},
'children': [],
'etype': 'task_manager',
'has': [],
'uid': 'task_manager.0000'
},
'wfprocessor.0000': {
'cfg': {},
'children': [],
'etype': 'wfprocessor',
'has': [],
'uid': 'wfprocessor.0000'
}
}
}
shutil.rmtree(amgr._sid)
def test_start_manager(self, mocked_init, mocked_Logger, mocked_Profiler):
rmq_params = mock.MagicMock(spec=ConnectionParameters)
rmgr = mock.MagicMock(spec=RPRmgr)
tmgr = RPTmgr('test_tmgr', ['pending_queues'], ['completed_queues'],
rmgr, rmq_params)
tmgr._log = mocked_Logger
tmgr._prof = mocked_Profiler
tmgr._uid = 'tmgr.0000'
tmgr._rmgr = 'test_rmgr'
tmgr._rmq_conn_params = rmq_params
tmgr._pending_queue = ['pending_queues']
tmgr._completed_queue = ['completed_queues']
tmgr._tmgr = _tmgr_side_effect
tmgr._tmgr_terminate = None
tmgr._tmgr_process = None
tmgr.start_manager()
try:
self.assertIsInstance(tmgr._tmgr_terminate, mp.synchronize.Event)
self.assertIsInstance(tmgr._tmgr_process, mp.context.Process)
pid = tmgr._tmgr_process.pid
self.assertTrue(psutil.pid_exists(pid))
finally:
if tmgr._tmgr_process.is_alive():
tmgr._tmgr_process.join()