async def job_resources(request, ws, username, project_name, experiment_sequence, job_sequence):
project = _get_project(username, project_name)
if not has_project_permissions(request.app.user, project, 'GET'):
exceptions.Forbidden("You don't have access to this project")
experiment = _get_validated_experiment(project, experiment_sequence)
job = _get_job(experiment, job_sequence)
job_uuid = job.uuid.hex
job_name = '{}.{}'.format(job.role, job.sequence)
if not RedisToStream.is_monitored_job_resources(job_uuid=job_uuid):
logger.info(
'Job resources with uuid `{}` is now being monitored'.format(job_name))
RedisToStream.monitor_job_resources(job_uuid=job_uuid)
if job_uuid in request.app.job_resources_ws_mangers:
ws_manager = request.app.job_resources_ws_mangers[job_uuid]
else:
ws_manager = SocketManager()
request.app.job_resources_ws_mangers[job_uuid] = ws_manager
def handle_job_disconnected_ws(ws):
ws_manager.remove_sockets(ws)
if len(ws_manager.ws) == 0:
logger.info('Stopping resources monitor for job {}'.format(job_name))
RedisToStream.remove_job_resources(job_uuid=job_uuid)
request.app.job_resources_ws_mangers.pop(job_uuid, None)
logger.info('Quitting resources socket for job {}'.format(job_name))
ws_manager.add_socket(ws)
should_check = 0
while True:
resources = RedisToStream.get_latest_job_resources(job=job_uuid, job_name=job_name)
should_check += 1
# After trying a couple of time, we must check the status of the job
if should_check > RESOURCES_CHECK:
job.refresh_from_db()
if job.is_done:
logger.info('removing all socket because the job `{}` is done'.format(job_name))
ws_manager.ws = set([])
handle_job_disconnected_ws(ws)
return
else:
should_check -= CHECK_DELAY
if resources:
try:
await ws.send(resources)
except ConnectionClosed:
handle_job_disconnected_ws(ws)
return
# Just to check if connection closed
if ws._connection_lost:
handle_job_disconnected_ws(ws)
return
await asyncio.sleep(SOCKET_SLEEP)
async def experiment_logs(request, ws, username, project_name,
experiment_sequence):
project = _get_project(username, project_name)
if not has_project_permissions(request.app.user, project, 'GET'):
exceptions.Forbidden("You don't have access to this project")
experiment = _get_validated_experiment(project, experiment_sequence)
experiment_uuid = experiment.uuid.hex
auditor.record(event_type=EXPERIMENT_LOGS_VIEWED,
instance=experiment,
actor_id=request.app.user.id)
if not RedisToStream.is_monitored_experiment_logs(
experiment_uuid=experiment_uuid):
logger.info('Experiment uuid `%s` logs is now being monitored',
experiment_uuid)
RedisToStream.monitor_experiment_logs(experiment_uuid=experiment_uuid)
# start consumer
if experiment_uuid in request.app.experiment_logs_consumers:
consumer = request.app.experiment_logs_consumers[experiment_uuid]
else:
logger.info('Add experiment log consumer for %s', experiment_uuid)
consumer = Consumer(
routing_key='{}.{}.*'.format(RoutingKeys.LOGS_SIDECARS,
experiment_uuid),
queue='{}.{}'.format(CeleryQueues.STREAM_LOGS_SIDECARS,
experiment_uuid))
request.app.experiment_logs_consumers[experiment_uuid] = consumer
consumer.run()
# add socket manager
consumer.add_socket(ws)
should_quite = False
num_message_retries = 0
while True:
num_message_retries += 1
for message in consumer.get_messages():
num_message_retries = 0
disconnected_ws = set()
for _ws in consumer.ws:
try:
await _ws.send(message)
except ConnectionClosed:
disconnected_ws.add(_ws)
consumer.remove_sockets(disconnected_ws)
# After trying a couple of time, we must check the status of the experiment
if num_message_retries > MAX_RETRIES:
experiment.refresh_from_db()
if experiment.is_done:
logger.info(
'removing all socket because the experiment `%s` is done',
experiment_uuid)
consumer.ws = set([])
else:
num_message_retries -= CHECK_DELAY
# Just to check if connection closed
if ws._connection_lost: # pylint:disable=protected-access
logger.info('Quitting logs socket for experiment uuid %s',
experiment_uuid)
consumer.remove_sockets({
ws,
})
should_quite = True
if not consumer.ws:
logger.info('Stopping logs monitor for experiment uuid %s',
experiment_uuid)
RedisToStream.remove_experiment_logs(
experiment_uuid=experiment_uuid)
# if experiment_uuid in request.app.experiment_logs_consumers:
# consumer = request.app.experiment_logs_consumers.pop(experiment_uuid, None)
# if consumer:
# consumer.stop()
should_quite = True
if should_quite:
return
await asyncio.sleep(SOCKET_SLEEP)
async def experiment_resources(request, ws, username, project_name,
experiment_sequence):
project = _get_project(username, project_name)
if not has_project_permissions(request.app.user, project, 'GET'):
exceptions.Forbidden("You don't have access to this project")
experiment = _get_validated_experiment(project, experiment_sequence)
experiment_uuid = experiment.uuid.hex
auditor.record(event_type=EXPERIMENT_RESOURCES_VIEWED,
instance=experiment,
actor_id=request.app.user.id)
if not RedisToStream.is_monitored_experiment_resources(
experiment_uuid=experiment_uuid):
logger.info(
'Experiment resource with uuid `%s` is now being monitored',
experiment_uuid)
RedisToStream.monitor_experiment_resources(
experiment_uuid=experiment_uuid)
if experiment_uuid in request.app.experiment_resources_ws_mangers:
ws_manager = request.app.experiment_resources_ws_mangers[
experiment_uuid]
else:
ws_manager = SocketManager()
request.app.experiment_resources_ws_mangers[
experiment_uuid] = ws_manager
def handle_experiment_disconnected_ws(ws):
ws_manager.remove_sockets(ws)
if not ws_manager.ws:
logger.info('Stopping resources monitor for uuid %s',
experiment_uuid)
RedisToStream.remove_experiment_resources(
experiment_uuid=experiment_uuid)
request.app.experiment_resources_ws_mangers.pop(
experiment_uuid, None)
logger.info('Quitting resources socket for uuid %s', experiment_uuid)
jobs = []
for job in experiment.jobs.values('uuid', 'role', 'sequence'):
job['uuid'] = job['uuid'].hex
job['name'] = '{}.{}'.format(job.pop('role'), job.pop('sequence'))
jobs.append(job)
ws_manager.add_socket(ws)
should_check = 0
while True:
resources = RedisToStream.get_latest_experiment_resources(jobs)
should_check += 1
# After trying a couple of time, we must check the status of the experiment
if should_check > RESOURCES_CHECK:
experiment.refresh_from_db()
if experiment.is_done:
logger.info(
'removing all socket because the experiment `%s` is done',
experiment_uuid)
ws_manager.ws = set([])
handle_experiment_disconnected_ws(ws)
return
else:
should_check -= CHECK_DELAY
if resources:
try:
await ws.send(resources)
except ConnectionClosed:
handle_experiment_disconnected_ws(ws)
return
# Just to check if connection closed
if ws._connection_lost: # pylint:disable=protected-access
handle_experiment_disconnected_ws(ws)
return
await asyncio.sleep(SOCKET_SLEEP)
async def job_logs(request, ws, username, project_name, experiment_sequence,
job_sequence):
project = _get_project(username, project_name)
if not has_project_permissions(request.app.user, project, 'GET'):
exceptions.Forbidden("You don't have access to this project")
experiment = _get_validated_experiment(project, experiment_sequence)
job = _get_job(experiment, job_sequence)
job_uuid = job.uuid.hex
if not RedisToStream.is_monitored_job_logs(job_uuid=job_uuid):
logger.info(
'Job uuid `{}` logs is now being monitored'.format(job_uuid))
RedisToStream.monitor_job_logs(job_uuid=job_uuid)
# start consumer
if job_uuid in request.app.job_logs_consumers:
consumer = request.app.job_logs_consumers[job_uuid]
else:
logger.info('Add job log consumer for {}'.format(job_uuid))
consumer = Consumer(
routing_key='{}.{}.{}'.format(RoutingKeys.LOGS_SIDECARS,
experiment.uuid.hex, job_uuid),
queue='{}.{}'.format(CeleryQueues.STREAM_LOGS_SIDECARS, job_uuid))
request.app.job_logs_consumers[job_uuid] = consumer
consumer.run()
# add socket manager
consumer.add_socket(ws)
should_quite = False
num_message_retries = 0
while True:
num_message_retries += 1
for message in consumer.get_messages():
num_message_retries = 0
disconnected_ws = set()
for _ws in consumer.ws:
try:
await _ws.send(message)
except ConnectionClosed:
disconnected_ws.add(_ws)
consumer.remove_sockets(disconnected_ws)
# After trying a couple of time, we must check the status of the experiment
if num_message_retries > MAX_RETRIES:
job.refresh_from_db()
if job.is_done:
logger.info(
'removing all socket because the job `{}` is done'.format(
job_uuid))
consumer.ws = set([])
else:
num_message_retries -= CHECK_DELAY
# Just to check if connection closed
if ws._connection_lost:
logger.info(
'Quitting logs socket for job uuid {}'.format(job_uuid))
consumer.remove_sockets({
ws,
})
should_quite = True
if len(consumer.ws) == 0:
logger.info(
'Stopping logs monitor for job uuid {}'.format(job_uuid))
RedisToStream.remove_job_logs(job_uuid=job_uuid)
# if job_uuid in request.app.job_logs_consumers:
# consumer = request.app.job_logs_consumers.pop(job_uuid, None)
# if consumer:
# consumer.stop()
should_quite = True
if should_quite:
return
await asyncio.sleep(SOCKET_SLEEP)