X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/f6071ef7bc7f6b7308c202e330cabd4ca111aadd..a464c2a216412032b58688194b4ff730fbe3a28f:/services/nodemanager/arvnodeman/launcher.py

diff --git a/services/nodemanager/arvnodeman/launcher.py b/services/nodemanager/arvnodeman/launcher.py
index 9c45318fdc..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
 
@@ -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)
@@ -66,25 +71,33 @@ 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,
                             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')
+    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 = {}
@@ -95,6 +108,7 @@ 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()
@@ -102,7 +116,7 @@ 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)
 
@@ -111,15 +125,17 @@ 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__)
+        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,
@@ -132,9 +148,10 @@ 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()
 
-        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,