def process_segment_rows(halt_event,
zeromq_context,
args,
node_dict,
node_databases,
raw_segment_rows):
"""
process handoffs of segment rows
"""
log = logging.getLogger("process_segment_rows")
rep_socket_uri = ipc_socket_uri(_socket_dir,
args.node_name,
"handoff_client")
prepare_ipc_path(rep_socket_uri)
rep_socket = zeromq_context.socket(zmq.REP)
rep_socket.setsockopt(zmq.SNDHWM, _socket_high_water_mark)
rep_socket.setsockopt(zmq.RCVHWM, _socket_high_water_mark)
log.info("binding rep socket to {0}".format(rep_socket_uri))
rep_socket.bind(rep_socket_uri)
log.debug("starting workers")
workers = list()
for index in range(args.worker_count):
worker_id = str(index+1)
workers.append(_start_worker_process(worker_id, args, rep_socket_uri))
# loop until all handoffs have been accomplished
log.debug("start handoffs")
work_generator = _generate_segment_rows(raw_segment_rows)
pending_handoff_count = 0
while not halt_event.is_set():
# get a segment row to process. If we are at EOF, segment_row = None
try:
source_node_names, segment_row = next(work_generator)
except StopIteration:
if pending_handoff_count == 0:
break
else:
source_node_names, segment_row = None, None
# if we have a segment row, and it is a tombstone, we can act
# directly on the node database(s) without sending it to a worker
if segment_row is not None:
if segment_row["status"] == segment_status_tombstone:
_process_tombstone(node_databases,
source_node_names,
segment_row)
_purge_handoff_from_source_nodes(node_databases,
source_node_names,
segment_row["collection_id"],
segment_row["key"],
segment_row["unified_id"],
segment_row["conjoined_part"],
segment_row["handoff_node_id"],
segment_status_tombstone)
continue
assert segment_row["status"] == segment_status_final, \
segment_row["status"]
# at this point we eaither have a segment row in final status, or
# None, indicating no more data
# block until we have a ready worker
try:
request = rep_socket.recv_pyobj()
except zmq.ZMQError as zmq_error:
if is_interrupted_system_call(zmq_error) and halt_event.is_set():
log.warn("breaking due to halt_event")
break
raise
assert not rep_socket.rcvmore
# see how the worker handled the previous segment (if any)
initial_request = False
if request["message-type"] == "start":
log.info("{0} initial request".format(request["worker-id"]))
initial_request = True
elif request["handoff-successful"]:
log.info("{0} handoff ({1}, {2}) successful".format(
request["worker-id"],
request["unified-id"],
request["conjoined-part"]))
assert pending_handoff_count > 0
pending_handoff_count -= 1
_purge_handoff_from_source_nodes(node_databases,
request["source-node-names"],
request["collection-id"],
request["key"],
request["unified-id"],
request["conjoined-part"],
request["handoff-node-id"],
segment_status_final)
else:
log.error("{0} handoff ({1}, {2}) failed: {3}".format(
request["worker-id"],
request["unified-id"],
request["conjoined-part"],
request["error-message"]))
assert pending_handoff_count > 0
pending_handoff_count -= 1
if segment_row is None:
# if we have no more work, tell the worker to stop
work_message = {"message-type" : "stop"}
else:
# otherwise, send the segment to the worker
work_message = {"message-type" : "work",
"source-node-names" : source_node_names,
"segment-row" : segment_row}
# if this is the worker's first request, send him the node_dict
if initial_request:
work_message["node-dict"] = node_dict
pending_handoff_count += 1
rep_socket.send_pyobj(work_message)
log.debug("end of handoffs")
for worker in workers:
terminate_subprocess(worker)
rep_socket.close()
def main():
"""
main entry point
returns 0 for normal termination (usually SIGTERM)
"""
return_value = 0
log_path = _log_path_template.format(os.environ["NIMBUSIO_LOG_DIR"],
_local_node_name)
initialize_logging(log_path)
log = logging.getLogger("main")
log.info("program starts")
halt_event = Event()
set_signal_handler(halt_event)
zeromq_context = zmq.Context()
resources = \
_resources_tuple(halt_event=Event(),
volume_by_space_id=_volume_name_by_space_id(),
pull_socket=zeromq_context.socket(zmq.PULL),
router_socket=zeromq_context.socket(zmq.ROUTER),
event_push_client=\
EventPushClient(zeromq_context,
"rs_io_controller"),
pending_work_by_volume=defaultdict(deque),
available_ident_by_volume=defaultdict(deque))
log.debug("binding to {0}".format(io_controller_pull_socket_uri))
resources.pull_socket.bind(io_controller_pull_socket_uri)
resources.router_socket.setsockopt(zmq.LINGER, 1000)
log.debug("binding to {0}".format(io_controller_router_socket_uri))
resources.router_socket.bind(io_controller_router_socket_uri)
# we poll the sockets for readability, we assume we can always
# write to the router socket
poller = zmq.Poller()
poller.register(resources.pull_socket, zmq.POLLIN | zmq.POLLERR)
poller.register(resources.router_socket, zmq.POLLIN | zmq.POLLERR)
worker_processes = list()
for volume_name in set(resources.volume_by_space_id.values()):
for index in range(_worker_count):
worker_processes.append(_launch_io_worker(volume_name, index + 1))
last_report_time = 0.0
try:
while not halt_event.is_set():
for worker_process in worker_processes:
poll_subprocess(worker_process)
for active_socket, event_flags in poller.poll(_poll_timeout):
if event_flags & zmq.POLLERR:
error_message = \
"error flags from zmq {0}".format(active_socket)
log.error(error_message)
raise PollError(error_message)
if active_socket is resources.pull_socket:
_read_pull_socket(resources)
elif active_socket is resources.router_socket:
_read_router_socket(resources)
else:
log.error("unknown socket {0}".format(active_socket))
current_time = time.time()
elapsed_time = current_time - last_report_time
if elapsed_time > _reporting_interval:
pending_work = 0
for volume_queue in resources.pending_work_by_volume.values():
pending_work += len(volume_queue)
report_message = \
"{0:,} pending_work entries".format(pending_work)
log.info(report_message)
resources.event_push_client.info("queue_sizes",
report_message,
pending_work=pending_work)
last_report_time = current_time
except zmq.ZMQError as zmq_error:
if is_interrupted_system_call(zmq_error) and halt_event.is_set():
log.info("program teminates normally with interrupted system call")
else:
log.exception("zeromq error processing request")
resources.event_push_client.exception(unhandled_exception_topic,
"zeromq_error",
exctype="ZMQError")
return_value = 1
except Exception as instance:
log.exception("error processing request")
resources.event_push_client.exception(
unhandled_exception_topic,
str(instance),
exctype=instance.__class__.__name__)
return_value = 1
else:
log.info("program teminates normally")
finally:
for worker_process in worker_processes:
terminate_subprocess(worker_process)
resources.pull_socket.close()
resources.router_socket.close()
resources.event_push_client.close()
zeromq_context.term()
return return_value
def main():
"""
main entry point
returns 0 for normal termination (usually SIGTERM)
"""
return_value = 0
log_path = _log_path_template.format(os.environ["NIMBUSIO_LOG_DIR"],
_local_node_name)
initialize_logging(log_path)
log = logging.getLogger("main")
log.info("program starts")
halt_event = Event()
set_signal_handler(halt_event)
zeromq_context = zmq.Context()
resources = \
_resources_tuple(halt_event=Event(),
zeromq_context=zeromq_context,
reply_push_sockets=dict(),
pull_socket=zeromq_context.socket(zmq.PULL),
io_controller_push_socket=\
zeromq_context.socket(zmq.PUSH),
router_socket=zeromq_context.socket(zmq.ROUTER),
event_push_client=\
EventPushClient(zeromq_context,
"rs_db_pool_controller"),
active_retrieves=dict(),
pending_work_queue=deque(),
available_ident_queue=deque())
log.debug("binding to {0}".format(db_controller_pull_socket_uri))
resources.pull_socket.bind(db_controller_pull_socket_uri)
log.debug("connecting to {0}".format(io_controller_pull_socket_uri))
resources.io_controller_push_socket.connect(io_controller_pull_socket_uri)
resources.router_socket.setsockopt(zmq.LINGER, 1000)
log.debug("binding to {0}".format(db_controller_router_socket_uri))
resources.router_socket.bind(db_controller_router_socket_uri)
# we poll the sockets for readability, we assume we can always
# write to the router socket
poller = zmq.Poller()
poller.register(resources.pull_socket, zmq.POLLIN | zmq.POLLERR)
poller.register(resources.router_socket, zmq.POLLIN| zmq.POLLERR)
worker_processes = list()
for index in range(_worker_count):
worker_processes.append(_launch_database_pool_worker(index+1))
last_report_time = 0.0
try:
while not halt_event.is_set():
for worker_process in worker_processes:
poll_subprocess(worker_process)
for active_socket, event_flags in poller.poll(_poll_timeout):
if event_flags & zmq.POLLERR:
error_message = \
"error flags from zmq {0}".format(active_socket)
log.error(error_message)
raise PollError(error_message)
if active_socket is resources.pull_socket:
_read_pull_socket(resources)
elif active_socket is resources.router_socket:
_read_router_socket(resources)
else:
log.error("unknown socket {0}".format(active_socket))
current_time = time.time()
elapsed_time = current_time - last_report_time
if elapsed_time > _reporting_interval:
report_message = \
"{0:,} active_retrives, " \
"{1:,} pending_work_queue entries, " \
"{2:,} available_ident_queue entries" \
"".format(len(resources.active_retrieves),
len(resources.pending_work_queue),
len(resources.available_ident_queue))
log.info(report_message)
resources.event_push_client.info(
"queue_sizes",
report_message,
active_retrieves=len(resources.active_retrieves),
pending_work_queue=len(resources.pending_work_queue),
available_ident_queue=len(resources.available_ident_queue))
last_report_time = current_time
except zmq.ZMQError as zmq_error:
if is_interrupted_system_call(zmq_error) and halt_event.is_set():
log.info("program teminates normally with interrupted system call")
else:
log.exception("zeromq error processing request")
resources.event_push_client.exception(unhandled_exception_topic,
"zeromq_error",
exctype="ZMQError")
return_value = 1
except Exception as instance:
log.exception("error processing request")
resources.event_push_client.exception(unhandled_exception_topic,
str(instance),
exctype=instance.__class__.__name__)
return_value = 1
else:
log.info("program teminates normally")
finally:
for worker_process in worker_processes:
terminate_subprocess(worker_process)
resources.pull_socket.close()
resources.io_controller_push_socket.close()
resources.router_socket.close()
for push_socket in resources.reply_push_sockets.values():
push_socket.close()
resources.event_push_client.close()
zeromq_context.term()
return return_value
def main():
"""
main entry point
returns 0 for normal termination (usually SIGTERM)
"""
return_value = 0
log_path = _log_path_template.format(os.environ["NIMBUSIO_LOG_DIR"], _local_node_name)
initialize_logging(log_path)
log = logging.getLogger("main")
log.info("program starts")
for internal_socket_uri in internal_socket_uri_list:
prepare_ipc_path(internal_socket_uri)
halt_event = Event()
set_signal_handler(halt_event)
database_pool_controller = _launch_database_pool_controller()
io_controller = _launch_io_controller()
zeromq_context = zmq.Context()
rep_socket = _bind_rep_socket(zeromq_context)
db_controller_push_socket = _connect_db_controller_push_socket(zeromq_context)
event_push_client = EventPushClient(zeromq_context, "retrieve_source")
event_push_client.info("program-starts", "retrieve source starts")
# we poll the sockets for readability, we assume we can always
# write to the push client sockets
poller = zmq.Poller()
poller.register(rep_socket, zmq.POLLIN | zmq.POLLERR)
last_report_time = 0.0
request_count = 0
try:
while not halt_event.is_set():
poll_subprocess(database_pool_controller)
poll_subprocess(io_controller)
# we've only registered one socket, so we could use an 'if' here,
# but this 'for' works ok and it has the same form as the other
# places where we use poller
for active_socket, event_flags in poller.poll(_poll_timeout):
if event_flags & zmq.POLLERR:
error_message = "error flags from zmq {0}".format(active_socket)
log.error(error_message)
raise PollError(error_message)
assert active_socket is rep_socket
_process_one_request(rep_socket, db_controller_push_socket)
request_count += 1
current_time = time.time()
elapsed_time = current_time - last_report_time
if elapsed_time > _reporting_interval:
report_message = "{0:,} requests".format(request_count)
log.info(report_message)
event_push_client.info("request_count", report_message, request_count=request_count)
last_report_time = current_time
request_count = 0
except KeyboardInterrupt: # convenience for testing
log.info("keyboard interrupt: terminating normally")
except zmq.ZMQError as zmq_error:
if is_interrupted_system_call(zmq_error) and halt_event.is_set():
log.info("program teminates normally with interrupted system call")
else:
log.exception("zeromq error processing request")
event_push_client.exception(unhandled_exception_topic, "zeromq_error", exctype="ZMQError")
return_value = 1
except Exception as instance:
log.exception("error processing request")
event_push_client.exception(unhandled_exception_topic, str(instance), exctype=instance.__class__.__name__)
return_value = 1
else:
log.info("program teminates normally")
finally:
terminate_subprocess(database_pool_controller)
terminate_subprocess(io_controller)
rep_socket.close()
db_controller_push_socket.close()
event_push_client.close()
zeromq_context.term()
return return_value
def process_segment_rows(halt_event, zeromq_context, args, node_dict,
node_databases, raw_segment_rows):
"""
process handoffs of segment rows
"""
log = logging.getLogger("process_segment_rows")
rep_socket_uri = ipc_socket_uri(_socket_dir, args.node_name,
"handoff_client")
prepare_ipc_path(rep_socket_uri)
rep_socket = zeromq_context.socket(zmq.REP)
rep_socket.setsockopt(zmq.SNDHWM, _socket_high_water_mark)
rep_socket.setsockopt(zmq.RCVHWM, _socket_high_water_mark)
log.info("binding rep socket to {0}".format(rep_socket_uri))
rep_socket.bind(rep_socket_uri)
log.debug("starting workers")
workers = list()
for index in range(args.worker_count):
worker_id = str(index + 1)
workers.append(_start_worker_process(worker_id, args, rep_socket_uri))
# loop until all handoffs have been accomplished
log.debug("start handoffs")
work_generator = _generate_segment_rows(raw_segment_rows)
pending_handoff_count = 0
while not halt_event.is_set():
# get a segment row to process. If we are at EOF, segment_row = None
try:
source_node_names, segment_row = next(work_generator)
except StopIteration:
if pending_handoff_count == 0:
break
else:
source_node_names, segment_row = None, None
# if we have a segment row, and it is a tombstone, we can act
# directly on the node database(s) without sending it to a worker
if segment_row is not None:
if segment_row["status"] == segment_status_tombstone:
_process_tombstone(node_databases, source_node_names,
segment_row)
_purge_handoff_from_source_nodes(
node_databases, source_node_names,
segment_row["collection_id"], segment_row["key"],
segment_row["unified_id"], segment_row["conjoined_part"],
segment_row["handoff_node_id"], segment_status_tombstone)
continue
assert segment_row["status"] == segment_status_final, \
segment_row["status"]
# at this point we eaither have a segment row in final status, or
# None, indicating no more data
# block until we have a ready worker
try:
request = rep_socket.recv_pyobj()
except zmq.ZMQError as zmq_error:
if is_interrupted_system_call(zmq_error) and halt_event.is_set():
log.warn("breaking due to halt_event")
break
raise
assert not rep_socket.rcvmore
# see how the worker handled the previous segment (if any)
initial_request = False
if request["message-type"] == "start":
log.info("{0} initial request".format(request["worker-id"]))
initial_request = True
elif request["handoff-successful"]:
log.info("{0} handoff ({1}, {2}) successful".format(
request["worker-id"], request["unified-id"],
request["conjoined-part"]))
assert pending_handoff_count > 0
pending_handoff_count -= 1
_purge_handoff_from_source_nodes(
node_databases, request["source-node-names"],
request["collection-id"], request["key"],
request["unified-id"], request["conjoined-part"],
request["handoff-node-id"], segment_status_final)
else:
log.error("{0} handoff ({1}, {2}) failed: {3}".format(
request["worker-id"], request["unified-id"],
request["conjoined-part"], request["error-message"]))
assert pending_handoff_count > 0
pending_handoff_count -= 1
if segment_row is None:
# if we have no more work, tell the worker to stop
work_message = {"message-type": "stop"}
else:
# otherwise, send the segment to the worker
work_message = {
"message-type": "work",
"source-node-names": source_node_names,
"segment-row": segment_row
}
# if this is the worker's first request, send him the node_dict
if initial_request:
work_message["node-dict"] = node_dict
pending_handoff_count += 1
rep_socket.send_pyobj(work_message)
log.debug("end of handoffs")
for worker in workers:
terminate_subprocess(worker)
rep_socket.close()
def main():
"""
main entry point
returns 0 for normal termination (usually SIGTERM)
"""
return_value = 0
log_path = _log_path_template.format(os.environ["NIMBUSIO_LOG_DIR"],
_local_node_name)
initialize_logging(log_path)
log = logging.getLogger("main")
log.info("program starts")
for internal_socket_uri in internal_socket_uri_list:
prepare_ipc_path(internal_socket_uri)
halt_event = Event()
set_signal_handler(halt_event)
database_pool_controller = _launch_database_pool_controller()
io_controller = _launch_io_controller()
zeromq_context = zmq.Context()
rep_socket = _bind_rep_socket(zeromq_context)
db_controller_push_socket = \
_connect_db_controller_push_socket(zeromq_context)
event_push_client = EventPushClient(zeromq_context, "retrieve_source")
event_push_client.info("program-starts", "retrieve source starts")
# we poll the sockets for readability, we assume we can always
# write to the push client sockets
poller = zmq.Poller()
poller.register(rep_socket, zmq.POLLIN | zmq.POLLERR)
last_report_time = 0.0
request_count = 0
try:
while not halt_event.is_set():
poll_subprocess(database_pool_controller)
poll_subprocess(io_controller)
# we've only registered one socket, so we could use an 'if' here,
# but this 'for' works ok and it has the same form as the other
# places where we use poller
for active_socket, event_flags in poller.poll(_poll_timeout):
if event_flags & zmq.POLLERR:
error_message = \
"error flags from zmq {0}".format(active_socket)
log.error(error_message)
raise PollError(error_message)
assert active_socket is rep_socket
_process_one_request(rep_socket, db_controller_push_socket)
request_count += 1
current_time = time.time()
elapsed_time = current_time - last_report_time
if elapsed_time > _reporting_interval:
report_message = "{0:,} requests".format(request_count)
log.info(report_message)
event_push_client.info("request_count",
report_message,
request_count=request_count)
last_report_time = current_time
request_count = 0
except KeyboardInterrupt: # convenience for testing
log.info("keyboard interrupt: terminating normally")
except zmq.ZMQError as zmq_error:
if is_interrupted_system_call(zmq_error) and halt_event.is_set():
log.info("program teminates normally with interrupted system call")
else:
log.exception("zeromq error processing request")
event_push_client.exception(unhandled_exception_topic,
"zeromq_error",
exctype="ZMQError")
return_value = 1
except Exception as instance:
log.exception("error processing request")
event_push_client.exception(unhandled_exception_topic,
str(instance),
exctype=instance.__class__.__name__)
return_value = 1
else:
log.info("program teminates normally")
finally:
terminate_subprocess(database_pool_controller)
terminate_subprocess(io_controller)
rep_socket.close()
db_controller_push_socket.close()
event_push_client.close()
zeromq_context.term()
return return_value
