def compss_persistent_executor(config):
# type: (PiperWorkerConfiguration) -> None
""" Persistent executor main function.
Retrieves the initial configuration and spawns the worker processes.
:param config: Piper Worker Configuration description.
:return: None
"""
COMM.gather(str(os.getpid()), root=0)
# Catch SIGTERM sent by bindings_piper
signal.signal(signal.SIGTERM, shutdown_handler)
# Catch SIGUSER2 to solve strange behaviour with mpi4py
signal.signal(signal.SIGUSR2, user_signal_handler)
# Set the binding in worker mode
import pycompss.util.context as context
context.set_pycompss_context(context.WORKER)
persistent_storage = (config.storage_conf != "null")
logger, logger_cfg, storage_loggers, _ = load_loggers(
config.debug, persistent_storage)
cache_profiler = False
if config.cache_profiler.lower() == 'true':
cache_profiler = True
if persistent_storage:
# Initialize storage
with event_worker(INIT_STORAGE_AT_WORKER_EVENT):
from storage.api import initWorker as initStorageAtWorker # noqa
initStorageAtWorker(config_file_path=config.storage_conf)
process_name = "".join(("Rank-", str(RANK)))
conf = ExecutorConf(config.debug, get_temporary_directory(), TRACING,
config.storage_conf, logger, logger_cfg,
persistent_storage, storage_loggers,
config.stream_backend, config.stream_master_name,
config.stream_master_port, CACHE_IDS, CACHE_QUEUE,
cache_profiler)
executor(None, process_name, config.pipes[RANK - 1], conf)
if persistent_storage:
# Finish storage
if __debug__:
logger.debug(HEADER + "Stopping persistent storage")
with event_worker(FINISH_STORAGE_AT_WORKER_EVENT):
from storage.api import finishWorker as finishStorageAtWorker # noqa
finishStorageAtWorker()
def compss_persistent_executor(config):
"""
Persistent worker main function.
Retrieves the initial configuration and spawns the worker processes.
:param config: Piper Worker Configuration description
:return: None
"""
import os
COMM.gather(str(os.getpid()), root=0)
# Catch SIGTERM sent by bindings_piper
signal.signal(signal.SIGTERM, shutdown_handler)
# Catch SIGUSER2 to solve strange behaviour with mpi4py
signal.signal(signal.SIGUSR2, user_signal_handler)
# Set the binding in worker mode
import pycompss.util.context as context
context.set_pycompss_context(context.WORKER)
persistent_storage = (config.storage_conf != 'null')
logger, storage_loggers = load_loggers(config.debug, persistent_storage,
config.tracing)
if persistent_storage:
# Initialize storage
with event(INIT_STORAGE_AT_WORKER_EVENT):
from storage.api import initWorker as initStorageAtWorker
initStorageAtWorker(config_file_path=config.storage_conf)
process_name = 'Rank-' + str(RANK)
conf = ExecutorConf(TRACING, config.storage_conf, logger, storage_loggers,
config.stream_backend, config.stream_master_name,
config.stream_master_port)
executor(None, process_name, config.pipes[RANK - 1], conf)
if persistent_storage:
# Finish storage
if __debug__:
logger.debug(HEADER + "Stopping persistent storage")
with event(FINISH_STORAGE_AT_WORKER_EVENT):
from storage.api import finishWorker as finishStorageAtWorker
finishStorageAtWorker()
def main():
# Emit sync event if tracing is enabled
tracing = sys.argv[1] == 'true'
task_id = int(sys.argv[2])
log_level = sys.argv[3]
storage_conf = sys.argv[4]
stream_backend = sys.argv[5]
stream_master_name = sys.argv[6]
stream_master_port = sys.argv[7]
# method_type = sys.argv[8]
params = sys.argv[9:]
# class_name = sys.argv[9]
# method_name = sys.argv[10]
# num_slaves = sys.argv[11]
# i = 11 + num_slaves
# slaves = sys.argv[11..i]
# numCus = sys.argv[i+1]
# has_target = sys.argv[i+2] == 'true'
# num_params = int(sys.argv[i+3])
# params = sys.argv[i+4..]
print("tracing = " + str(tracing))
print("task_id = " + str(task_id))
print("log_level = " + str(log_level))
print("storage_conf = " + str(storage_conf))
persistent_storage = False
if storage_conf != 'null':
persistent_storage = True
streaming = False
if stream_backend not in [None, 'null', 'NONE']:
streaming = True
with trace_multiprocessing_worker() if tracing else dummy_context():
if streaming:
# Start streaming
DistroStreamClientHandler.init_and_start(
master_ip=stream_master_name, master_port=stream_master_port)
# Load log level configuration file
worker_path = os.path.dirname(os.path.realpath(__file__))
if log_level == 'true' or log_level == "debug":
# Debug
init_logging_worker(
worker_path + '/../../../log/logging_gat_worker_debug.json',
tracing)
elif log_level == "info" or log_level == "off":
# Info or no debug
init_logging_worker(
worker_path + '/../../../log/logging_gat_worker_off.json',
tracing)
else:
# Default
init_logging_worker(
worker_path + '/../../../log/logging_gat_worker.json', tracing)
if persistent_storage:
# Initialize storage
with event(INIT_STORAGE_AT_WORKER_EVENT):
from storage.api import initWorker as initStorageAtWorker
initStorageAtWorker(config_file_path=storage_conf)
# Init worker
exit_code = compss_worker(tracing, str(task_id), storage_conf, params)
if streaming:
# Finish streaming
DistroStreamClientHandler.set_stop()
if persistent_storage:
# Finish storage
with event(FINISH_STORAGE_AT_WORKER_EVENT):
from storage.api import finishWorker as finishStorageAtWorker
finishStorageAtWorker()
if exit_code == 1:
exit(1)
def compss_persistent_worker():
# Get args
debug = (sys.argv[1] == 'true')
tracing = (sys.argv[2] == 'true')
storage_conf = sys.argv[3]
tasks_x_node = int(sys.argv[4])
in_pipes = sys.argv[5:5 + tasks_x_node]
out_pipes = sys.argv[5 + tasks_x_node:]
if tracing:
import pyextrae.multiprocessing as pyextrae
pyextrae.eventandcounters(SYNC_EVENTS, 1)
pyextrae.eventandcounters(TASK_EVENTS, WORKER_RUNNING)
if debug:
assert tasks_x_node == len(in_pipes)
assert tasks_x_node == len(out_pipes)
persistent_storage = False
if storage_conf != 'null':
persistent_storage = True
from storage.api import initWorker as initStorageAtWorker
from storage.api import finishWorker as finishStorageAtWorker
# Load log level configuration file
worker_path = os.path.dirname(os.path.realpath(__file__))
if debug:
# Debug
init_logging_worker(worker_path + '/../../log/logging.json.debug')
else:
# Default
init_logging_worker(worker_path + '/../../log/logging.json.off')
if __debug__:
logger = logging.getLogger('pycompss.worker.worker')
logger.debug("[PYTHON WORKER] piper_worker.py wake up")
logger.debug("[PYTHON WORKER] -----------------------------")
logger.debug("[PYTHON WORKER] Persistent worker parameters:")
logger.debug("[PYTHON WORKER] -----------------------------")
logger.debug("[PYTHON WORKER] Debug : " + str(debug))
logger.debug("[PYTHON WORKER] Tracing : " + str(tracing))
logger.debug("[PYTHON WORKER] Tasks per node : " + str(tasks_x_node))
logger.debug("[PYTHON WORKER] In Pipes : " + str(in_pipes))
logger.debug("[PYTHON WORKER] Out Pipes : " + str(out_pipes))
logger.debug("[PYTHON WORKER] Storage conf. : " + str(storage_conf))
logger.debug("[PYTHON WORKER] -----------------------------")
if persistent_storage:
# Initialize storage
initStorageAtWorker(config_file_path=storage_conf)
# Create new threads
queues = []
for i in range(0, tasks_x_node):
if __debug__:
logger.debug("[PYTHON WORKER] Launching process " + str(i))
process_name = 'Process-' + str(i)
queues.append(Queue())
def create_threads():
processes.append(Process(target=worker, args=(queues[i],
process_name,
in_pipes[i],
out_pipes[i],
storage_conf)))
processes[i].start()
create_threads()
# Catch SIGTERM send by bindings_piper to exit all processes
signal.signal(signal.SIGTERM, shutdown_handler)
# Wait for all threads
for i in range(0, tasks_x_node):
processes[i].join()
# Check if there is any exception message from the threads
for i in range(0, tasks_x_node):
if not queues[i].empty:
print(queues[i].get())
for q in queues:
q.close()
q.join_thread()
if persistent_storage:
# Finish storage
finishStorageAtWorker()
if __debug__:
logger.debug("[PYTHON WORKER] Finished")
if tracing:
pyextrae.eventandcounters(TASK_EVENTS, 0)
pyextrae.eventandcounters(SYNC_EVENTS, 0)
# pyextrae.eventandcounters(TASK_EVENTS, 0)
pyextrae.eventandcounters(TASK_EVENTS, WORKER_INITIALIZATION)
# Load log level configuration file
worker_path = os.path.dirname(os.path.realpath(__file__))
if log_level == 'true' or log_level == "debug":
# Debug
init_logging_worker(worker_path + '/../../log/logging.json.debug')
elif log_level == "info" or log_level == "off":
# Info or no debug
init_logging_worker(worker_path + '/../../log/logging.json.off')
else:
# Default
init_logging_worker(worker_path + '/../../log/logging.json')
if persistent_storage:
# Initialize storage
initStorageAtWorker(config_file_path=storage_conf)
# Init worker
compss_worker(persistent_storage)
if tracing:
pyextrae.eventandcounters(TASK_EVENTS, 0)
# pyextrae.eventandcounters(TASK_EVENTS, PROCESS_DESTRUCTION)
pyextrae.eventandcounters(SYNC_EVENTS, taskId)
if persistent_storage:
# Finish storage
finishStorageAtWorker()
def main():
# Emit sync event if tracing is enabled
tracing = sys.argv[1] == 'true'
task_id = int(sys.argv[2])
log_level = sys.argv[3]
storage_conf = sys.argv[4]
stream_backend = sys.argv[5]
stream_master_name = sys.argv[6]
stream_master_port = sys.argv[7]
# method_type = sys.argv[8]
params = sys.argv[9:]
# class_name = sys.argv[9]
# method_name = sys.argv[10]
# num_slaves = sys.argv[11]
# i = 11 + num_slaves
# slaves = sys.argv[11..i]
# numCus = sys.argv[i+1]
# has_target = sys.argv[i+2] == 'true'
# num_params = int(sys.argv[i+3])
# params = sys.argv[i+4..]
print("tracing = " + str(tracing))
print("task_id = " + str(task_id))
print("log_level = " + str(log_level))
print("storage_conf = " + str(storage_conf))
persistent_storage = False
if storage_conf != 'null':
persistent_storage = True
from storage.api import initWorker as initStorageAtWorker
from storage.api import finishWorker as finishStorageAtWorker
streaming = False
if stream_backend not in [None, 'null', 'NONE']:
streaming = True
if tracing:
# Start tracing
import pyextrae.multiprocessing as pyextrae
pyextrae.eventandcounters(SYNC_EVENTS, task_id)
# pyextrae.eventandcounters(TASK_EVENTS, 0)
pyextrae.eventandcounters(TASK_EVENTS, WORKER_RUNNING_EVENT)
if streaming:
# Start streaming
DistroStreamClientHandler.init_and_start(
master_ip=stream_master_name, master_port=stream_master_port)
# Load log level configuration file
worker_path = os.path.dirname(os.path.realpath(__file__))
if log_level == 'true' or log_level == "debug":
# Debug
init_logging_worker(worker_path + '/../../../log/logging_debug.json')
elif log_level == "info" or log_level == "off":
# Info or no debug
init_logging_worker(worker_path + '/../../../log/logging_off.json')
else:
# Default
init_logging_worker(worker_path + '/../../../log/logging.json')
if persistent_storage:
# Initialize storage
initStorageAtWorker(config_file_path=storage_conf)
# Init worker
exit_code = compss_worker(tracing, str(task_id), storage_conf, params)
if tracing:
# Finish tracing
pyextrae.eventandcounters(TASK_EVENTS, 0)
# pyextrae.eventandcounters(TASK_EVENTS, PROCESS_DESTRUCTION)
pyextrae.eventandcounters(SYNC_EVENTS, task_id)
if streaming:
# Finish streaming
DistroStreamClientHandler.set_stop()
if persistent_storage:
# Finish storage
finishStorageAtWorker()
if exit_code == 1:
exit(1)
def compss_persistent_worker(config):
# type: (PiperWorkerConfiguration) -> None
""" Persistent worker main function.
Retrieves the initial configuration and spawns the worker processes.
:param config: Piper Worker Configuration description.
:return: None
"""
pids = COMM.gather(str(os.getpid()), root=0)
# Catch SIGTERM sent by bindings_piper
signal.signal(signal.SIGTERM, shutdown_handler)
# Catch SIGUSER2 to solve strange behaviour with mpi4py
signal.signal(signal.SIGUSR2, user_signal_handler)
# Set the binding in worker mode
import pycompss.util.context as context
context.set_pycompss_context(context.WORKER)
persistent_storage = (config.storage_conf != 'null')
logger, _, _ = load_loggers(config.debug, persistent_storage)
if __debug__:
logger.debug(HEADER + "mpi_piper_worker.py rank: " + str(RANK) +
" wake up")
config.print_on_logger(logger)
# Start storage
if persistent_storage:
# Initialize storage
if __debug__:
logger.debug(HEADER + "Starting persistent storage")
from storage.api import initWorker as initStorageAtWorker # noqa
initStorageAtWorker(config_file_path=config.storage_conf)
for i in range(0, config.tasks_x_node):
child_in_pipe = config.pipes[i].input_pipe
try:
child_pid = pids[i + 1]
except IndexError:
child_pid = pids[i]
PROCESSES[child_in_pipe] = child_pid
if __debug__:
logger.debug(HEADER + "Starting alive")
logger.debug(HEADER + "Control pipe: " + str(config.control_pipe))
# Read command from control pipe
alive = True
control_pipe = config.control_pipe
while alive:
command = control_pipe.read_command()
if command != "":
line = command.split()
if line[0] == ADD_EXECUTOR_TAG:
in_pipe = line[1]
out_pipe = line[2]
control_pipe.write(" ".join(
(ADD_EXECUTOR_FAILED_TAG, out_pipe, in_pipe, str(0))))
elif line[0] == REMOVE_EXECUTOR_TAG:
in_pipe = line[1]
out_pipe = line[2]
PROCESSES.pop(in_pipe, None)
control_pipe.write(" ".join(
(REMOVED_EXECUTOR_TAG, out_pipe, in_pipe)))
elif line[0] == QUERY_EXECUTOR_ID_TAG:
in_pipe = line[1]
out_pipe = line[2]
pid = PROCESSES.get(in_pipe)
control_pipe.write(" ".join(
(REPLY_EXECUTOR_ID_TAG, out_pipe, in_pipe, str(pid))))
elif line[0] == CANCEL_TASK_TAG:
in_pipe = line[1]
pid = PROCESSES.get(in_pipe)
logger.debug(HEADER + "Signaling process with PID " + pid +
" to cancel a task")
os.kill(
int(pid),
signal.SIGUSR2) # NOSONAR cancellation produced by COMPSs
elif line[0] == PING_TAG:
control_pipe.write(PONG_TAG)
elif line[0] == QUIT_TAG:
alive = False
else:
logger.debug(HEADER + "ERROR: UNKNOWN COMMAND: " + command)
alive = False
# Stop storage
if persistent_storage:
# Finish storage
logger.debug(HEADER + "Stopping persistent storage")
from storage.api import finishWorker as finishStorageAtWorker # noqa
finishStorageAtWorker()
if __debug__:
logger.debug(HEADER + "Finished")
control_pipe.write(QUIT_TAG)
control_pipe.close()
def main():
# type: () -> None
""" GAT worker main code.
Executes the task provided by parameters.
:return: None
"""
# Emit sync event if tracing is enabled
tracing = sys.argv[1] == 'true'
task_id = int(sys.argv[2])
log_level = sys.argv[3]
storage_conf = sys.argv[4]
stream_backend = sys.argv[5]
stream_master_name = sys.argv[6]
stream_master_port = sys.argv[7]
# Next: method_type = sys.argv[8]
params = sys.argv[9:]
# Next parameters:
# class_name = sys.argv[10]
# method_name = sys.argv[11]
# num_slaves = sys.argv[12]
# i = 13 + num_slaves
# slaves = sys.argv[12..i]
# numCus = sys.argv[i+1]
# has_target = sys.argv[i+2] == 'true'
# num_params = int(sys.argv[i+3])
# params = sys.argv[i+4..]
if log_level == "true" or log_level == "debug":
print("Tracing = " + str(tracing))
print("Task id = " + str(task_id))
print("Log level = " + str(log_level))
print("Storage conf = " + str(storage_conf))
persistent_storage = False
if storage_conf != "null":
persistent_storage = True
streaming = False
if stream_backend not in [None, "null", "NONE"]:
streaming = True
with trace_multiprocessing_worker() if tracing else dummy_context():
if streaming:
# Start streaming
DistroStreamClientHandler.init_and_start(
master_ip=stream_master_name, master_port=stream_master_port)
# Load log level configuration file
worker_path = os.path.dirname(os.path.realpath(__file__))
if log_level == "true" or log_level == "debug":
# Debug
log_json = "".join(
(worker_path, "/../../../log/logging_gat_worker_debug.json"))
elif log_level == "info" or log_level == "off":
# Info or no debug
log_json = "".join(
(worker_path, "/../../../log/logging_gat_worker_off.json"))
else:
# Default
log_json = "".join(
(worker_path, "/../../../log/logging_gat_worker.json"))
init_logging_worker(log_json, tracing)
if persistent_storage:
# Initialize storage
with event_worker(INIT_STORAGE_AT_WORKER_EVENT):
from storage.api import initWorker as initStorageAtWorker # noqa
initStorageAtWorker(config_file_path=storage_conf)
# Init worker
exit_code = compss_worker(tracing, str(task_id), storage_conf, params,
log_json)
if streaming:
# Finish streaming
DistroStreamClientHandler.set_stop()
if persistent_storage:
# Finish storage
with event_worker(FINISH_STORAGE_AT_WORKER_EVENT):
from storage.api import finishWorker as finishStorageAtWorker # noqa
finishStorageAtWorker()
if exit_code == 1:
exit(1)
def compss_persistent_worker(config):
# type: (PiperWorkerConfiguration) -> None
""" Persistent worker main function.
Retrieves the initial configuration and spawns the worker processes.
:param config: Piper Worker Configuration description.
:return: None
"""
global CACHE
global CACHE_PROCESS
# Catch SIGTERM sent by bindings_piper
signal.signal(signal.SIGTERM, shutdown_handler)
# Set the binding in worker mode
context.set_pycompss_context(context.WORKER)
persistent_storage = (config.storage_conf != 'null')
logger, logger_cfg, storage_loggers, log_dir = load_loggers(config.debug, persistent_storage)
if __debug__:
logger.debug(HEADER + "piper_worker.py wake up")
config.print_on_logger(logger)
if persistent_storage:
# Initialize storage
logger.debug(HEADER + "Starting persistent storage")
with event_worker(INIT_STORAGE_AT_WORKER_EVENT):
from storage.api import initWorker as initStorageAtWorker # noqa
initStorageAtWorker(config_file_path=config.storage_conf)
# Create new processes
queues = []
cache_profiler = False
if config.cache_profiler.lower() == 'true':
cache_profiler = True
# Setup cache
if is_cache_enabled(str(config.cache)):
# Deploy the necessary processes
CACHE = True
cache_params = start_cache(logger, str(config.cache), cache_profiler, log_dir)
else:
# No cache
CACHE = False
cache_params = (None, None, None, None) # type: ignore
smm, CACHE_PROCESS, cache_queue, cache_ids = cache_params
# Create new executor processes
conf = ExecutorConf(config.debug,
get_temporary_directory(),
TRACING,
config.storage_conf,
logger,
logger_cfg,
persistent_storage,
storage_loggers,
config.stream_backend,
config.stream_master_name,
config.stream_master_port,
cache_ids,
cache_queue,
cache_profiler)
for i in range(0, config.tasks_x_node):
if __debug__:
logger.debug(HEADER + "Launching process " + str(i))
process_name = "".join(("Process-", str(i)))
pid, queue = create_executor_process(process_name, conf, config.pipes[i])
queues.append(queue)
# Read command from control pipe
alive = True
process_counter = config.tasks_x_node
control_pipe = config.control_pipe # type: typing.Any
while alive:
command = control_pipe.read_command(retry_period=1)
if command != "":
line = command.split()
if line[0] == ADD_EXECUTOR_TAG:
process_name = "".join(("Process-", str(process_counter)))
process_counter = process_counter + 1
in_pipe = line[1]
out_pipe = line[2]
pipe = Pipe(in_pipe, out_pipe)
pid, queue = create_executor_process(process_name, conf, pipe)
queues.append(queue)
control_pipe.write(" ".join((ADDED_EXECUTOR_TAG,
out_pipe,
in_pipe,
str(pid))))
elif line[0] == QUERY_EXECUTOR_ID_TAG:
in_pipe = line[1]
out_pipe = line[2]
proc = PROCESSES.get(in_pipe) # type: typing.Any
pid = proc.pid
control_pipe.write(" ".join((REPLY_EXECUTOR_ID_TAG,
out_pipe,
in_pipe,
str(pid))))
elif line[0] == CANCEL_TASK_TAG:
in_pipe = line[1]
cancel_proc = PROCESSES.get(in_pipe) # type: typing.Any
cancel_pid = cancel_proc.pid
if __debug__:
logger.debug(HEADER + "Signaling process with PID " +
str(cancel_pid) + " to cancel a task")
os.kill(cancel_pid, signal.SIGUSR2) # NOSONAR cancellation produced by COMPSs
elif line[0] == REMOVE_EXECUTOR_TAG:
in_pipe = line[1]
out_pipe = line[2]
proc = PROCESSES.pop(in_pipe, None)
if proc:
if proc.is_alive():
logger.warn(HEADER + "Forcing terminate on : " +
proc.name)
proc.terminate()
proc.join()
control_pipe.write(" ".join((REMOVED_EXECUTOR_TAG,
out_pipe,
in_pipe)))
elif line[0] == PING_TAG:
control_pipe.write(PONG_TAG)
elif line[0] == QUIT_TAG:
alive = False
# Wait for all threads
for proc in PROCESSES.values():
proc.join()
# Check if there is any exception message from the threads
for i in range(0, config.tasks_x_node):
if not queues[i].empty:
logger.error(HEADER + "Exception in threads queue: " +
str(queues[i].get()))
for queue in queues:
queue.close()
queue.join_thread()
if CACHE:
stop_cache(smm, cache_queue, cache_profiler, CACHE_PROCESS) # noqa
if persistent_storage:
# Finish storage
if __debug__:
logger.debug(HEADER + "Stopping persistent storage")
with event_worker(FINISH_STORAGE_AT_WORKER_EVENT):
from storage.api import finishWorker as finishStorageAtWorker # noqa
finishStorageAtWorker()
if __debug__:
logger.debug(HEADER + "Finished")
control_pipe.write(QUIT_TAG)
control_pipe.close()
def compss_persistent_worker(config):
"""
Persistent worker main function.
Retrieves the initial configuration and spawns the worker processes.
:param config: Piper Worker Configuration description
:return: None
"""
# Catch SIGTERM sent by bindings_piper
signal.signal(signal.SIGTERM, shutdown_handler)
# Set the binding in worker mode
context.set_pycompss_context(context.WORKER)
if TRACING:
try:
user_paths = os.environ['PYTHONPATH']
except KeyError:
user_paths = ""
print("PYTHON PATH = " + user_paths)
import pyextrae.multiprocessing as pyextrae
pyextrae.eventandcounters(SYNC_EVENTS, 1)
pyextrae.eventandcounters(TASK_EVENTS, WORKER_RUNNING_EVENT)
persistent_storage = (config.storage_conf != 'null')
logger, storage_loggers = load_loggers(config.debug, persistent_storage)
if __debug__:
logger.debug(HEADER + "piper_worker.py wake up")
config.print_on_logger(logger)
if persistent_storage:
# Initialize storage
logger.debug(HEADER + "Starting persitent storage")
from storage.api import initWorker as initStorageAtWorker
initStorageAtWorker(config_file_path=config.storage_conf)
# Create new threads
queues = []
for i in range(0, config.tasks_x_node):
if __debug__:
logger.debug(HEADER + "Launching process " + str(i))
process_name = 'Process-' + str(i)
queue = Queue()
queues.append(queue)
conf = ExecutorConf(TRACING,
config.storage_conf,
logger,
storage_loggers,
config.stream_backend,
config.stream_master_name,
config.stream_master_port)
process = Process(target=executor, args=(queue,
process_name,
config.pipes[i],
conf))
PROCESSES[config.pipes[i].input_pipe] = process
process.start()
# Read command from control pipe
alive = True
process_counter = config.tasks_x_node
control_pipe = config.control_pipe
while alive:
command = control_pipe.read_command(retry_period=1)
if command != "":
line = command.split()
if line[0] == ADD_EXECUTOR_TAG:
process_name = 'Process-' + str(process_counter)
process_counter = process_counter + 1
in_pipe = line[1]
out_pipe = line[2]
pipe = Pipe(in_pipe, out_pipe)
pid = create_threads(process_name, pipe)
control_pipe.write(ADDED_EXECUTOR_TAG + " " +
out_pipe + " " +
in_pipe + " " +
str(pid))
elif line[0] == QUERY_EXECUTOR_ID_TAG:
in_pipe = line[1]
out_pipe = line[2]
proc = PROCESSES.get(in_pipe)
pid = proc.pid
control_pipe.write(REPLY_EXECUTOR_ID_TAG + " " +
out_pipe + " " +
in_pipe + " " +
str(pid))
elif line[0] == CANCEL_TASK_TAG:
in_pipe = line[1]
proc = PROCESSES.get(in_pipe)
pid = proc.pid
logger.debug("[PYTHON WORKER] Signaling process with PID " + str(pid) + " to cancel a task")
kill(pid, signal.SIGUSR2)
elif line[0] == REMOVE_EXECUTOR_TAG:
in_pipe = line[1]
out_pipe = line[2]
proc = PROCESSES.pop(in_pipe, None)
if proc:
if proc.is_alive():
logger.warn(HEADER + "Forcing terminate on : " +
proc.name)
proc.terminate()
proc.join()
control_pipe.write(REMOVED_EXECUTOR_TAG + " " +
out_pipe + " " +
in_pipe)
elif line[0] == PING_TAG:
control_pipe.write(PONG_TAG)
elif line[0] == QUIT_TAG:
alive = False
# Wait for all threads
for proc in PROCESSES.values():
proc.join()
# Check if there is any exception message from the threads
for i in range(0, config.tasks_x_node):
if not queues[i].empty:
logger.error(HEADER + "Exception in threads queue: " +
str(queues[i].get()))
for queue in queues:
queue.close()
queue.join_thread()
if persistent_storage:
# Finish storage
logger.debug(HEADER + "Stopping persistent storage")
from storage.api import finishWorker as finishStorageAtWorker
finishStorageAtWorker()
if __debug__:
logger.debug(HEADER + "Finished")
if TRACING:
pyextrae.eventandcounters(TASK_EVENTS, 0)
pyextrae.eventandcounters(SYNC_EVENTS, 0)
control_pipe.write(QUIT_TAG)
control_pipe.close()
