X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/09794d996eca79b85d3ac0c21a4a43c65a51d0d7..59a972c831bc8f7cd4e896ed8e1c71277b97f04e:/services/nodemanager/arvnodeman/launcher.py diff --git a/services/nodemanager/arvnodeman/launcher.py b/services/nodemanager/arvnodeman/launcher.py index cb80bbf293..34ea9adb3d 100644 --- a/services/nodemanager/arvnodeman/launcher.py +++ b/services/nodemanager/arvnodeman/launcher.py @@ -1,4 +1,7 @@ #!/usr/bin/env python +# Copyright (C) The Arvados Authors. All rights reserved. +# +# SPDX-License-Identifier: AGPL-3.0 from __future__ import absolute_import, print_function @@ -68,7 +71,7 @@ def setup_logging(path, level, **sublevels): return root_logger def build_server_calculator(config): - cloud_size_list = config.node_sizes(config.new_cloud_client().list_sizes()) + cloud_size_list = config.node_sizes() if not cloud_size_list: abort("No valid node sizes configured") return ServerCalculator(cloud_size_list, @@ -77,17 +80,24 @@ def build_server_calculator(config): config.getfloat('Daemon', 'node_mem_scaling')) def launch_pollers(config, server_calculator): - poll_time = config.getint('Daemon', 'poll_time') + poll_time = config.getfloat('Daemon', 'poll_time') max_poll_time = config.getint('Daemon', 'max_poll_time') + cloudlist_poll_time = config.getfloat('Daemon', 'cloudlist_poll_time') or poll_time + nodelist_poll_time = config.getfloat('Daemon', 'nodelist_poll_time') or poll_time + wishlist_poll_time = config.getfloat('Daemon', 'wishlist_poll_time') or poll_time + timer = TimedCallBackActor.start(poll_time / 10.0).tell_proxy() cloud_node_poller = CloudNodeListMonitorActor.start( - config.new_cloud_client(), timer, server_calculator, poll_time, max_poll_time).tell_proxy() + config.new_cloud_client(), timer, server_calculator, cloudlist_poll_time, max_poll_time).tell_proxy() arvados_node_poller = ArvadosNodeListMonitorActor.start( - config.new_arvados_client(), timer, poll_time, max_poll_time).tell_proxy() + config.new_arvados_client(), timer, nodelist_poll_time, max_poll_time).tell_proxy() job_queue_poller = JobQueueMonitorActor.start( config.new_arvados_client(), timer, server_calculator, - poll_time, max_poll_time).tell_proxy() + config.getboolean('Arvados', 'jobs_queue'), + config.getboolean('Arvados', 'slurm_queue'), + wishlist_poll_time, max_poll_time + ).tell_proxy() return timer, cloud_node_poller, arvados_node_poller, job_queue_poller _caught_signals = {} @@ -119,13 +129,13 @@ def main(args=None): try: root_logger = setup_logging(config.get('Logging', 'file'), **config.log_levels()) - root_logger.info("%s %s, libcloud %s", sys.argv[0], __version__, libcloud.__version__) + root_logger.info("%s %s started, libcloud %s", sys.argv[0], __version__, libcloud.__version__) node_setup, node_shutdown, node_update, node_monitor = \ config.dispatch_classes() server_calculator = build_server_calculator(config) timer, cloud_node_poller, arvados_node_poller, job_queue_poller = \ launch_pollers(config, server_calculator) - cloud_node_updater = node_update.start(config.new_cloud_client).tell_proxy() + cloud_node_updater = node_update.start(config.new_cloud_client, timer).tell_proxy() node_daemon = NodeManagerDaemonActor.start( job_queue_poller, arvados_node_poller, cloud_node_poller, cloud_node_updater, timer, @@ -138,7 +148,8 @@ def main(args=None): config.getint('Daemon', 'boot_fail_after'), config.getint('Daemon', 'node_stale_after'), node_setup, node_shutdown, node_monitor, - max_total_price=config.getfloat('Daemon', 'max_total_price')).tell_proxy() + max_total_price=config.getfloat('Daemon', 'max_total_price'), + consecutive_idle_count=config.getint('Daemon', 'consecutive_idle_count'),).tell_proxy() watchdog = WatchdogActor.start(config.getint('Daemon', 'watchdog'), cloud_node_poller.actor_ref,