def test_monitor_experiment_resources(self): expeirment_uuid = uuid.uuid4().hex RedisToStream.monitor_experiment_resources(expeirment_uuid) assert RedisToStream.is_monitored_experiment_resources( expeirment_uuid) is True RedisToStream.remove_experiment_resources(expeirment_uuid) assert RedisToStream.is_monitored_experiment_resources( expeirment_uuid) is False
def test_experiment_monitoring(self): experiment_uuid = uuid.uuid4().hex assert RedisToStream.is_monitored_experiment_resources(experiment_uuid) is False RedisToStream.monitor_experiment_resources(experiment_uuid) assert RedisToStream.is_monitored_experiment_resources(experiment_uuid) is True RedisToStream.remove_experiment_resources(experiment_uuid) assert RedisToStream.is_monitored_experiment_resources(experiment_uuid) is False assert RedisToStream.is_monitored_experiment_logs(experiment_uuid) is False RedisToStream.monitor_experiment_logs(experiment_uuid) assert RedisToStream.is_monitored_experiment_logs(experiment_uuid) is True RedisToStream.remove_experiment_logs(experiment_uuid) assert RedisToStream.is_monitored_experiment_logs(experiment_uuid) is False
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 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)