#!/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
import pykka
import arvnodeman.daemon as nmdaemon
+import arvnodeman.status as status
from arvnodeman.jobqueue import ServerCalculator
from arvnodeman.computenode.dispatch import ComputeNodeMonitorActor
from . import testutil
+from . import test_status
import logging
class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
cloud_node = testutil.cloud_node_mock(4)
arv_node = testutil.arvados_node_mock(4, crunch_worker_state='down')
setup = self.start_node_boot(cloud_node, arv_node)
- self.daemon.node_up(setup).get(self.TIMEOUT)
+ self.daemon.node_setup_finished(setup).get(self.TIMEOUT)
self.assertEqual(1, self.alive_monitor_count())
self.daemon.update_arvados_nodes([arv_node])
self.daemon.update_cloud_nodes([cloud_node])
cloud_node = testutil.cloud_node_mock(1)
setup = self.start_node_boot(cloud_node)
self.daemon.update_cloud_nodes([cloud_node])
- self.daemon.node_up(setup).get(self.TIMEOUT)
+ self.daemon.node_setup_finished(setup).get(self.TIMEOUT)
self.assertEqual(1, self.alive_monitor_count())
def test_no_duplication_when_booted_node_listed(self):
cloud_node = testutil.cloud_node_mock(2)
setup = self.start_node_boot(cloud_node, id_num=2)
- self.daemon.node_up(setup)
+ self.daemon.node_setup_finished(setup)
self.daemon.update_cloud_nodes([cloud_node]).get(self.TIMEOUT)
self.assertEqual(1, self.alive_monitor_count())
# even it doesn't appear in the listing (e.g., because of delays
# propagating tags).
setup = self.start_node_boot()
- self.daemon.node_up(setup).get(self.TIMEOUT)
+ self.daemon.node_setup_finished(setup).get(self.TIMEOUT)
self.assertEqual(1, self.alive_monitor_count())
self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
self.assertEqual(1, self.alive_monitor_count())
def test_booted_unlisted_node_counted(self):
setup = self.start_node_boot(id_num=1)
- self.daemon.node_up(setup)
+ self.daemon.node_setup_finished(setup)
self.daemon.update_server_wishlist(
[testutil.MockSize(1)]).get(self.TIMEOUT)
self.stop_proxy(self.daemon)
def test_booted_node_can_shutdown(self):
setup = self.start_node_boot()
- self.daemon.node_up(setup).get(self.TIMEOUT)
+ self.daemon.node_setup_finished(setup).get(self.TIMEOUT)
self.assertEqual(1, self.alive_monitor_count())
monitor = self.monitor_list()[0].proxy()
self.daemon.update_server_wishlist([])
self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
+ self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
self.stop_proxy(self.daemon)
self.assertTrue(self.node_shutdown.start.called,
"daemon did not shut down booted node on offer")
+ with test_status.TestServer() as srv:
+ self.assertEqual(0, srv.get_status().get('nodes_unpaired', None))
+ self.assertEqual(1, srv.get_status().get('nodes_shutdown', None))
+ self.assertEqual(0, srv.get_status().get('nodes_wish', None))
+
def test_booted_node_lifecycle(self):
cloud_node = testutil.cloud_node_mock(6)
setup = self.start_node_boot(cloud_node, id_num=6)
- self.daemon.node_up(setup).get(self.TIMEOUT)
+ self.daemon.node_setup_finished(setup).get(self.TIMEOUT)
self.assertEqual(1, self.alive_monitor_count())
monitor = self.monitor_list()[0].proxy()
self.daemon.update_server_wishlist([])
def test_booted_node_shut_down_when_never_listed(self):
setup = self.start_node_boot()
self.cloud_factory().node_start_time.return_value = time.time() - 3601
- self.daemon.node_up(setup).get(self.TIMEOUT)
+ self.daemon.node_setup_finished(setup).get(self.TIMEOUT)
self.assertEqual(1, self.alive_monitor_count())
self.assertFalse(self.node_shutdown.start.called)
now = time.time()
cloud_node = testutil.cloud_node_mock(2)
setup = self.start_node_boot(cloud_node)
self.cloud_factory().node_start_time.return_value = time.time() - 3601
- self.daemon.node_up(setup).get(self.TIMEOUT)
+ self.daemon.node_setup_finished(setup).get(self.TIMEOUT)
self.assertEqual(1, self.alive_monitor_count())
self.daemon.update_cloud_nodes([cloud_node])
self.monitor_list()[0].tell_proxy().consider_shutdown()
arv_node = testutil.arvados_node_mock(4, crunch_worker_state='down')
setup = self.start_node_boot(cloud_node, arv_node)
self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
- self.daemon.node_up(setup).get(self.TIMEOUT)
+ self.daemon.node_setup_finished(setup).get(self.TIMEOUT)
self.assertEqual(1, self.alive_monitor_count())
self.monitor_list()[0].proxy().cloud_node_start_time = time.time()-3601
self.daemon.update_cloud_nodes([cloud_node])
cloud_node = testutil.cloud_node_mock(3)
arv_node = testutil.arvados_node_mock(3)
setup = self.start_node_boot(cloud_node, arv_node)
- self.daemon.node_up(setup).get(self.TIMEOUT)
+ self.daemon.node_setup_finished(setup).get(self.TIMEOUT)
self.assertEqual(1, self.alive_monitor_count())
self.daemon.update_cloud_nodes([cloud_node])
self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
cloud_node = testutil.cloud_node_mock(5)
arv_node = testutil.arvados_node_mock(5, job_uuid=True)
setup = self.start_node_boot(cloud_node, arv_node)
- self.daemon.node_up(setup).get(self.TIMEOUT)
+ self.daemon.node_setup_finished(setup).get(self.TIMEOUT)
self.assertEqual(1, self.alive_monitor_count())
self.daemon.update_cloud_nodes([cloud_node])
self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
self.daemon.node_finished_shutdown(self.last_shutdown).get(self.TIMEOUT)
self.assertEqual(0, self.alive_monitor_count())
- def test_broken_node_not_counted(self):
- size = testutil.MockSize(8)
- cloud_node = testutil.cloud_node_mock(8, size=size)
- wishlist = [size]
- self.make_daemon([cloud_node], [testutil.arvados_node_mock(8)],
- wishlist, avail_sizes=[(size, {"cores":1})])
- self.assertEqual(1, self.alive_monitor_count())
- self.assertFalse(self.node_setup.start.called)
- monitor = self.monitor_list()[0].proxy()
- shutdown_proxy = self.node_shutdown.start().proxy
- shutdown_proxy().cloud_node.get.return_value = cloud_node
- shutdown_proxy().success.get.return_value = False
- self.cloud_factory().broken.return_value = True
- self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
- self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
- self.daemon.node_finished_shutdown(shutdown_proxy()).get(self.TIMEOUT)
- self.daemon.update_cloud_nodes([cloud_node]).get(self.TIMEOUT)
- self.daemon.update_server_wishlist(wishlist).get(self.TIMEOUT)
- self.stop_proxy(self.daemon)
- self.assertEqual(1, self.node_setup.start.call_count)
-
def test_nodes_shutting_down_replaced_below_max_nodes(self):
size = testutil.MockSize(6)
cloud_node = testutil.cloud_node_mock(6, size=size)
new_node.stop_if_no_cloud_node.reset_mock()
self.daemon.shutdown().get(self.TIMEOUT)
self.assertTrue(new_node.stop_if_no_cloud_node.called)
- self.daemon.node_up(new_node).get(self.TIMEOUT)
+ self.daemon.node_setup_finished(new_node).get(self.TIMEOUT)
self.assertTrue(new_node.stop.called)
self.timer.deliver()
self.assertTrue(