12055: Avoid RAM exhaustion on bootup by asking AWS only the AMI

[arvados.git] / services / nodemanager / arvnodeman / launcher.py

diff --git a/services/nodemanager/arvnodeman/launcher.py b/services/nodemanager/arvnodeman/launcher.py
index 93f6cbdbe3c60c4736614594fdd2e77666276274..d85ef552c064ac0b9f4527c3b93194ede60c142f 100644 (file)
--- 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
 
@@ -13,6 +16,7 @@ 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
@@ -21,6 +25,7 @@ from .timedcallback import TimedCallBackActor
 from ._version import __version__
 
 node_daemon = None
+watchdog = None
 
 def abort(msg, code=1):
     print("arvados-node-manager: " + msg)
@@ -85,7 +90,10 @@ def launch_pollers(config, server_calculator):
         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).tell_proxy()
+        config.getboolean('Arvados', 'jobs_queue'),
+        config.getboolean('Arvados', 'slurm_queue'),
+        poll_time, max_poll_time
+    ).tell_proxy()
     return timer, cloud_node_poller, arvados_node_poller, job_queue_poller
 
 _caught_signals = {}
@@ -96,6 +104,7 @@ def shutdown_signal(signal_code, frame):
         pykka.ActorRegistry.stop_all()
         sys.exit(-signal_code)
     elif current_count == 0:
+        watchdog.stop()
         node_daemon.shutdown()
     elif current_count == 1:
         pykka.ActorRegistry.stop_all()
@@ -103,7 +112,7 @@ def shutdown_signal(signal_code, frame):
         sys.exit(-signal_code)
 
 def main(args=None):
-    global node_daemon
+    global node_daemon, watchdog
     args = parse_cli(args)
     config = load_config(args.config)
 
@@ -112,6 +121,8 @@ def main(args=None):
     for sigcode in [signal.SIGINT, signal.SIGQUIT, signal.SIGTERM]:
         signal.signal(sigcode, shutdown_signal)
 
+    status.Server(config).start()
+
     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__)
@@ -120,7 +131,7 @@ def main(args=None):
         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,
@@ -135,7 +146,7 @@ def main(args=None):
             node_setup, node_shutdown, node_monitor,
             max_total_price=config.getfloat('Daemon', 'max_total_price')).tell_proxy()
 
-        WatchdogActor.start(config.getint('Daemon', 'watchdog'),
+        watchdog = WatchdogActor.start(config.getint('Daemon', 'watchdog'),
                             cloud_node_poller.actor_ref,
                             arvados_node_poller.actor_ref,
                             job_queue_poller.actor_ref,