+ def test_node_shutdown_after_cancelled_shutdown(self):
+ cloud_node = testutil.cloud_node_mock(5)
+ self.make_daemon([cloud_node], [testutil.arvados_node_mock(5)])
+ self.assertEqual(1, self.alive_monitor_count())
+ monitor = self.monitor_list()[0].proxy()
+ self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
+ self.last_shutdown.success.get.return_value = False
+ self.daemon.node_finished_shutdown(self.last_shutdown).get(self.TIMEOUT)
+ self.assertEqual(1, self.alive_monitor_count())
+
+ self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
+ self.last_shutdown.success.get.return_value = True
+ self.last_shutdown.stop.side_effect = lambda: monitor.stop()
+ self.daemon.node_finished_shutdown(self.last_shutdown).get(self.TIMEOUT)
+ self.assertEqual(0, self.alive_monitor_count())
+
+ def test_broken_node_blackholed_after_cancelled_shutdown(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
+ shutdown_proxy().cancel_reason.get.return_value = self.node_shutdown.NODE_BROKEN
+ 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)
+ self.make_daemon([cloud_node], [testutil.arvados_node_mock(6)],
+ avail_sizes=[(size, {"cores":1})])
+ self.assertEqual(1, self.alive_monitor_count())
+ monitor = self.monitor_list()[0].proxy()
+ self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
+ self.assertTrue(self.node_shutdown.start.called)
+ self.daemon.update_server_wishlist(
+ [testutil.MockSize(6)]).get(self.TIMEOUT)
+ self.stop_proxy(self.daemon)
+ self.assertTrue(self.node_setup.start.called)
+
+ def test_nodes_shutting_down_not_replaced_at_max_nodes(self):
+ cloud_node = testutil.cloud_node_mock(7)
+ self.make_daemon([cloud_node], [testutil.arvados_node_mock(7)],
+ max_nodes=1)
+ self.assertEqual(1, self.alive_monitor_count())
+ monitor = self.monitor_list()[0].proxy()
+ self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
+ self.assertTrue(self.node_shutdown.start.called)
+ self.daemon.update_server_wishlist(
+ [testutil.MockSize(7)]).get(self.TIMEOUT)
+ self.stop_proxy(self.daemon)
+ self.assertFalse(self.node_setup.start.called)
+
+ def test_nodes_shutting_down_count_against_excess(self):
+ size = testutil.MockSize(8)
+ cloud_nodes = [testutil.cloud_node_mock(n, size=size) for n in [8, 9]]
+ arv_nodes = [testutil.arvados_node_mock(n, size=size) for n in [8, 9]]
+ self.make_daemon(cloud_nodes, arv_nodes, [size],
+ avail_sizes=[(size, {"cores":1})])
+ self.assertEqual(2, self.alive_monitor_count())
+ for mon_ref in self.monitor_list():
+ self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
+ self.assertEqual(1, self.node_shutdown.start.call_count)
+