def test_set_latest_job_resources(self):
gpu_resources = {
'index':
0,
'bus_id':
'0000:00:1E.1',
'memory_free':
1000,
'memory_total':
12883853312,
'memory_used':
8388608000,
'memory_utilization':
0,
'minor':
1,
'name':
'GeForce GTX TITAN 0',
'power_draw':
125,
'power_limit':
250,
'processes': [{
'command': 'python',
'gpu_memory_usage': 4000,
'pid': 48448,
'username': '******'
}, {
'command': 'python',
'gpu_memory_usage': 4000,
'pid': 153223,
'username': '******'
}],
'serial':
'0322917092147',
'temperature_gpu':
80,
'utilization_gpu':
76,
'uuid':
'GPU-10fb0fbd-2696-43f3-467f-d280d906a107'
}
config_dict = {
'job_uuid': uuid.uuid4().hex,
'experiment_uuid': uuid.uuid4().hex,
'container_id':
'3175e88873af9077688cee20eaadc0c07746efb84d01ae696d6d17ed9bcdfbc4',
'cpu_percentage': 0.6947691836734693,
'percpu_percentage': [0.4564075715616173, 0.23836161211185192],
'memory_used': 84467712,
'memory_limit': 2096160768,
'gpu_resources': gpu_resources
}
RedisToStream.set_latest_job_resources(config_dict['job_uuid'],
config_dict)
config_dict['job_name'] = 'master.0'
assert config_dict == RedisToStream.get_latest_job_resources(
config_dict['job_uuid'], 'master.0', True)
async def experiment_job_resources(request, ws, username, project_name,
experiment_id, job_id):
job, _, message = validate_experiment_job(request=request,
username=username,
project_name=project_name,
experiment_id=experiment_id,
job_id=job_id)
if job is None:
await ws.send(get_error_message(message))
return
job_uuid = job.uuid.hex
job_name = '{}.{}'.format(job.role, job.id)
auditor.record(event_type=EXPERIMENT_JOB_RESOURCES_VIEWED,
instance=job,
actor_id=request.app.user.id,
actor_name=request.app.user.username)
if not RedisToStream.is_monitored_job_resources(job_uuid=job_uuid):
logger.info('Job resources with uuid `%s` is now being monitored',
job_name)
RedisToStream.monitor_job_resources(job_uuid=job_uuid)
if job_uuid in request.app.job_resources_ws_managers:
ws_manager = request.app.job_resources_ws_managers[job_uuid]
else:
ws_manager = SocketManager()
request.app.job_resources_ws_managers[job_uuid] = ws_manager
def handle_job_disconnected_ws(ws):
ws_manager.remove_sockets(ws)
if not ws_manager.ws:
logger.info('Stopping resources monitor for job %s', job_name)
RedisToStream.remove_job_resources(job_uuid=job_uuid)
request.app.job_resources_ws_managers.pop(job_uuid, None)
logger.info('Quitting resources socket for job %s', 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 `%s` is done',
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: # pylint:disable=protected-access
handle_job_disconnected_ws(ws)
return
await asyncio.sleep(SOCKET_SLEEP)
