async def log_job(request, ws, job, pod_id, namespace, container):
job_uuid = job.uuid.hex
if job_uuid in request.app.job_logs_ws_managers:
ws_manager = request.app.job_logs_ws_managers[job_uuid]
else:
ws_manager = SocketManager()
request.app.job_logs_ws_managers[job_uuid] = ws_manager
ws_manager.add_socket(ws)
# Stream phase changes
status = None
while status != JobLifeCycle.RUNNING and not JobLifeCycle.is_done(status):
job.refresh_from_db()
if status != job.last_status:
status = job.last_status
await notify_ws(ws=ws, message=get_status_message(status))
if should_disconnect(ws=ws, ws_manager=ws_manager):
return
await asyncio.sleep(SOCKET_SLEEP)
if JobLifeCycle.is_done(status):
await notify_ws(ws=ws, message=get_status_message(status))
return
config.load_incluster_config()
k8s_api = client.CoreV1Api()
await log_job_pod(k8s_api=k8s_api,
ws=ws,
ws_manager=ws_manager,
pod_id=pod_id,
container=container,
namespace=namespace)
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)
auditor.record(event_type=EXPERIMENT_JOB_RESOURCES_VIEWED,
instance=job,
actor_id=request.app.user.id)
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_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 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_mangers.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)
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)
auditor.record(event_type=EXPERIMENT_JOB_RESOURCES_VIEWED,
instance=job,
actor_id=request.app.user.id)
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_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 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_mangers.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)
async def log_experiment(request, ws, experiment, namespace, container):
experiment_uuid = experiment.uuid.hex
if experiment_uuid in request.app.experiment_logs_ws_managers:
ws_manager = request.app.experiment_logs_ws_managers[experiment_uuid]
else:
ws_manager = SocketManager()
request.app.experiment_logs_ws_managers[experiment_uuid] = ws_manager
ws_manager.add_socket(ws)
# Stream phase changes
status = None
while status != ExperimentLifeCycle.RUNNING and not ExperimentLifeCycle.is_done(
status):
experiment.refresh_from_db()
if status != experiment.last_status:
status = experiment.last_status
await notify_ws(ws=ws, message=get_status_message(status))
if should_disconnect(ws=ws, ws_manager=ws_manager):
return
await asyncio.sleep(SOCKET_SLEEP)
if ExperimentLifeCycle.is_done(status):
await notify_ws(ws=ws, message=get_status_message(status))
return
config.load_incluster_config()
k8s_api = client.CoreV1Api()
log_requests = []
for job in experiment.jobs.all():
pod_id = job.pod_id
log_requests.append(
log_job_pod(k8s_api=k8s_api,
ws=ws,
ws_manager=ws_manager,
pod_id=pod_id,
container=container,
namespace=namespace,
task_type=job.role,
task_idx=job.sequence))
await asyncio.wait(log_requests)
async def experiment_resources(request, ws, username, project_name, experiment_id):
experiment, message = validate_experiment(request=request,
username=username,
project_name=project_name,
experiment_id=experiment_id)
if experiment is None:
await ws.send(get_error_message(message))
return
experiment_uuid = experiment.uuid.hex
auditor.record(event_type=EXPERIMENT_RESOURCES_VIEWED,
instance=experiment,
actor_id=request.app.user.id,
actor_name=request.app.user.username)
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_managers:
ws_manager = request.app.experiment_resources_ws_managers[experiment_uuid]
else:
ws_manager = SocketManager()
request.app.experiment_resources_ws_managers[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_managers.pop(experiment_uuid, None)
logger.info('Quitting resources socket for uuid %s', experiment_uuid)
jobs = []
for job in experiment.jobs.values('uuid', 'role', 'id'):
job['uuid'] = job['uuid'].hex
job['name'] = '{}.{}'.format(job.pop('role'), job.pop('id'))
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)