self.check_monitors_arvados_nodes(arv_node)
def test_arvados_node_un_and_re_paired(self):
+ # We need to create the Arvados node mock after spinning up the daemon
+ # to make sure it's new enough to pair with the cloud node.
+ self.make_daemon([testutil.cloud_node_mock(3)], arvados_nodes=None)
arv_node = testutil.arvados_node_mock(3)
- self.make_daemon([testutil.cloud_node_mock(3)], [arv_node])
+ self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
self.check_monitors_arvados_nodes(arv_node)
self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
self.assertEqual(0, self.alive_monitor_count())
self.make_daemon([testutil.cloud_node_mock()],
want_sizes=[testutil.MockSize(1)])
self.stop_proxy(self.daemon)
- self.assertFalse(self.node_setup.called)
+ self.assertFalse(self.node_setup.start.called)
+
+ def test_dont_count_missing_as_busy(self):
+ size = testutil.MockSize(1)
+ self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1),
+ testutil.cloud_node_mock(2)],
+ arvados_nodes=[testutil.arvados_node_mock(1),
+ testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
+ want_sizes=[size, size])
+ self.stop_proxy(self.daemon)
+ self.assertTrue(self.node_setup.start.called)
+
+ def test_missing_counts_towards_max(self):
+ size = testutil.MockSize(1)
+ self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1),
+ testutil.cloud_node_mock(2)],
+ arvados_nodes=[testutil.arvados_node_mock(1),
+ testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
+ want_sizes=[size, size],
+ max_nodes=2)
+ self.stop_proxy(self.daemon)
+ self.assertFalse(self.node_setup.start.called)
+
+ def test_excess_counts_missing(self):
+ size = testutil.MockSize(1)
+ cloud_nodes = [testutil.cloud_node_mock(1), testutil.cloud_node_mock(2)]
+ self.make_daemon(cloud_nodes=cloud_nodes,
+ arvados_nodes=[testutil.arvados_node_mock(1),
+ testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
+ want_sizes=[size])
+ 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)
+
+ def test_missing_shutdown_not_excess(self):
+ size = testutil.MockSize(1)
+ cloud_nodes = [testutil.cloud_node_mock(1), testutil.cloud_node_mock(2)]
+ self.make_daemon(cloud_nodes=cloud_nodes,
+ arvados_nodes=[testutil.arvados_node_mock(1),
+ testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
+ want_sizes=[size])
+ self.daemon.shutdowns.get()[cloud_nodes[1].id] = True
+ 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(0, self.node_shutdown.start.call_count)
def test_booting_nodes_counted(self):
cloud_node = testutil.cloud_node_mock(1)
self.last_setup.arvados_node.get.return_value = arv_node
return self.last_setup
+ def test_no_new_node_when_booted_node_not_usable(self):
+ 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.assertEqual(1, self.alive_monitor_count())
+ self.daemon.update_cloud_nodes([cloud_node])
+ self.daemon.update_arvados_nodes([arv_node])
+ self.daemon.update_server_wishlist(
+ [testutil.MockSize(1)]).get(self.TIMEOUT)
+ self.stop_proxy(self.daemon)
+ self.assertEqual(1, self.node_setup.start.call_count)
+
def test_no_duplication_when_booting_node_listed_fast(self):
# Test that we don't start two ComputeNodeMonitorActors when
# we learn about a booting node through a listing before we
self.stop_proxy(self.daemon)
self.assertShutdownCancellable(False)
+ def test_booted_node_shut_down_when_never_working(self):
+ 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.assertEqual(1, self.alive_monitor_count())
+ self.daemon.update_cloud_nodes([cloud_node])
+ self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
+ self.timer.deliver()
+ self.stop_proxy(self.daemon)
+ self.assertShutdownCancellable(False)
+
def test_node_that_pairs_not_considered_failed_boot(self):
cloud_node = testutil.cloud_node_mock(3)
arv_node = testutil.arvados_node_mock(3)
self.stop_proxy(self.daemon)
self.assertFalse(self.node_shutdown.start.called)
+ def test_node_that_pairs_busy_not_considered_failed_boot(self):
+ 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.assertEqual(1, self.alive_monitor_count())
+ self.daemon.update_cloud_nodes([cloud_node])
+ self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
+ self.timer.deliver()
+ self.stop_proxy(self.daemon)
+ self.assertFalse(self.node_shutdown.start.called)
+
def test_booting_nodes_shut_down(self):
self.make_daemon(want_sizes=[testutil.MockSize(1)])
self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
self.stop_proxy(self.daemon)
self.assertTrue(self.last_setup.stop_if_no_cloud_node.called)
+ def test_all_booting_nodes_tried_to_shut_down(self):
+ size = testutil.MockSize(2)
+ self.make_daemon(want_sizes=[size])
+ self.daemon.max_nodes.get(self.TIMEOUT)
+ setup1 = self.last_setup
+ setup1.stop_if_no_cloud_node().get.return_value = False
+ setup1.stop_if_no_cloud_node.reset_mock()
+ self.daemon.update_server_wishlist([size, size]).get(self.TIMEOUT)
+ self.daemon.max_nodes.get(self.TIMEOUT)
+ self.assertIsNot(setup1, self.last_setup)
+ self.last_setup.stop_if_no_cloud_node().get.return_value = True
+ self.last_setup.stop_if_no_cloud_node.reset_mock()
+ self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
+ self.daemon.max_nodes.get(self.TIMEOUT)
+ self.stop_proxy(self.daemon)
+ self.assertEqual(1, self.last_setup.stop_if_no_cloud_node.call_count)
+ self.assertTrue(setup1.stop_if_no_cloud_node.called)
+
def test_shutdown_declined_at_wishlist_capacity(self):
cloud_node = testutil.cloud_node_mock(1)
size = testutil.MockSize(1)
self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
self.assertTrue(shutdown_proxy.called)
+ def test_broken_node_blackholed_after_cancelled_shutdown(self):
+ cloud_node = testutil.cloud_node_mock(8)
+ wishlist = [testutil.MockSize(8)]
+ self.make_daemon([cloud_node], [testutil.arvados_node_mock(8)],
+ wishlist)
+ 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):
cloud_node = testutil.cloud_node_mock(6)
self.make_daemon([cloud_node], [testutil.arvados_node_mock(6)])
def test_clean_shutdown_waits_for_node_setup_finish(self):
new_node = self.start_node_boot()
+ new_node.stop_if_no_cloud_node().get.return_value = False
+ 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.actor_ref.actor_stopped.wait(self.TIMEOUT))
def test_wishlist_ignored_after_shutdown(self):
- size = testutil.MockSize(2)
- self.make_daemon(want_sizes=[size])
+ new_node = self.start_node_boot()
+ new_node.stop_if_no_cloud_node().get.return_value = False
+ new_node.stop_if_no_cloud_node.reset_mock()
self.daemon.shutdown().get(self.TIMEOUT)
+ size = testutil.MockSize(2)
self.daemon.update_server_wishlist([size] * 2).get(self.TIMEOUT)
self.timer.deliver()
self.stop_proxy(self.daemon)
self.assertEqual(1, self.node_setup.start.call_count)
+
+ def test_shutdown_actor_stopped_when_cloud_node_delisted(self):
+ self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
+ self.assertEqual(1, self.alive_monitor_count())
+ monitor = self.monitor_list()[0].proxy()
+ self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
+ self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
+ self.stop_proxy(self.daemon)
+ self.assertEqual(
+ 1, self.node_shutdown.start().proxy().stop().get.call_count)
+
+ def test_shutdown_actor_cleanup_copes_with_dead_actors(self):
+ self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
+ self.assertEqual(1, self.alive_monitor_count())
+ monitor = self.monitor_list()[0].proxy()
+ self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
+ # We're mainly testing that update_cloud_nodes catches and handles
+ # the ActorDeadError.
+ stop_method = self.node_shutdown.start().proxy().stop().get
+ stop_method.side_effect = pykka.ActorDeadError
+ self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
+ self.stop_proxy(self.daemon)
+ self.assertEqual(1, stop_method.call_count)
projects / arvados.git / blobdiff