import daemon
import pykka
+import libcloud
from . import config as nmconfig
+from . import status
+from .baseactor import WatchdogActor
from .daemon import NodeManagerDaemonActor
from .jobqueue import JobQueueMonitorActor, ServerCalculator
from .nodelist import ArvadosNodeListMonitorActor, CloudNodeListMonitorActor
from .timedcallback import TimedCallBackActor
+from ._version import __version__
node_daemon = None
parser = argparse.ArgumentParser(
prog='arvados-node-manager',
description="Dynamically allocate Arvados cloud compute nodes")
+ parser.add_argument(
+ '--version', action='version',
+ version="%s %s" % (sys.argv[0], __version__),
+ help='Print version and exit.')
parser.add_argument(
'--foreground', action='store_true', default=False,
help="Run in the foreground. Don't daemonize.")
for logger_name, sublevel in sublevels.iteritems():
sublogger = logging.getLogger(logger_name)
sublogger.setLevel(sublevel)
+ return root_logger
def build_server_calculator(config):
cloud_size_list = config.node_sizes(config.new_cloud_client().list_sizes())
abort("No valid node sizes configured")
return ServerCalculator(cloud_size_list,
config.getint('Daemon', 'max_nodes'),
- config.getfloat('Daemon', 'max_total_price'))
+ config.getfloat('Daemon', 'max_total_price'),
+ config.getfloat('Daemon', 'node_mem_scaling'))
def launch_pollers(config, server_calculator):
poll_time = config.getint('Daemon', 'poll_time')
max_poll_time = config.getint('Daemon', 'max_poll_time')
- timer = TimedCallBackActor.start(poll_time / 10.0).proxy()
+ timer = TimedCallBackActor.start(poll_time / 10.0).tell_proxy()
cloud_node_poller = CloudNodeListMonitorActor.start(
- config.new_cloud_client(), timer, poll_time, max_poll_time).proxy()
+ config.new_cloud_client(), timer, server_calculator, poll_time, max_poll_time).tell_proxy()
arvados_node_poller = ArvadosNodeListMonitorActor.start(
- config.new_arvados_client(), timer, poll_time, max_poll_time).proxy()
+ config.new_arvados_client(), timer, poll_time, max_poll_time).tell_proxy()
job_queue_poller = JobQueueMonitorActor.start(
config.new_arvados_client(), timer, server_calculator,
- poll_time, max_poll_time).proxy()
+ poll_time, max_poll_time).tell_proxy()
return timer, cloud_node_poller, arvados_node_poller, job_queue_poller
_caught_signals = {}
for sigcode in [signal.SIGINT, signal.SIGQUIT, signal.SIGTERM]:
signal.signal(sigcode, shutdown_signal)
+ status.Server(config).start()
+
try:
- setup_logging(config.get('Logging', 'file'), **config.log_levels())
+ root_logger = setup_logging(config.get('Logging', 'file'), **config.log_levels())
+ root_logger.info("%s %s, 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).proxy()
+ cloud_node_updater = node_update.start(config.new_cloud_client).tell_proxy()
node_daemon = NodeManagerDaemonActor.start(
job_queue_poller, arvados_node_poller, cloud_node_poller,
cloud_node_updater, timer,
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')).proxy()
+ max_total_price=config.getfloat('Daemon', 'max_total_price')).tell_proxy()
+
+ WatchdogActor.start(config.getint('Daemon', 'watchdog'),
+ cloud_node_poller.actor_ref,
+ arvados_node_poller.actor_ref,
+ job_queue_poller.actor_ref,
+ node_daemon.actor_ref)
signal.pause()
daemon_stopped = node_daemon.actor_ref.actor_stopped.is_set