class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
unittest.TestCase):
- def busywait(self, f, finalize=None):
+ def busywait(self, f):
n = 0
- while not f() and n < 20:
+ while not f() and n < 200:
time.sleep(.1)
self.daemon.ping().get(self.TIMEOUT)
n += 1
- if finalize is not None:
- finalize()
self.assertTrue(f())
def mock_node_start(self, **kwargs):
self.daemon.update_server_wishlist(want_sizes).get(self.TIMEOUT)
def monitor_list(self):
- return pykka.ActorRegistry.get_by_class(ComputeNodeMonitorActor)
+ return [c.actor.actor_ref for c in self.daemon.cloud_nodes.get(self.TIMEOUT).nodes.values() if c.actor]
- def monitored_arvados_nodes(self):
+ def monitored_arvados_nodes(self, include_unpaired=True):
pairings = []
for future in [actor.proxy().arvados_node
for actor in self.monitor_list()]:
try:
- pairings.append(future.get(self.TIMEOUT))
+ g = future.get(self.TIMEOUT)
+ if g or include_unpaired:
+ pairings.append(g)
except pykka.ActorDeadError:
pass
return pairings
def alive_monitor_count(self):
return len(self.monitored_arvados_nodes())
+ def paired_monitor_count(self):
+ return len(self.monitored_arvados_nodes(False))
+
def assertShutdownCancellable(self, expected=True):
self.assertTrue(self.node_shutdown.start.called)
self.assertIs(expected,
def test_easy_node_creation(self):
size = testutil.MockSize(1)
self.make_daemon(want_sizes=[size])
- self.busywait(lambda: self.node_setup.start.called,
- lambda: self.stop_proxy(self.daemon))
+ self.busywait(lambda: self.node_setup.start.called)
def check_monitors_arvados_nodes(self, *arv_nodes):
+ self.busywait(lambda: len(arv_nodes) == len(self.monitored_arvados_nodes()))
self.assertItemsEqual(arv_nodes, self.monitored_arvados_nodes())
def test_node_pairing(self):
cloud_node = testutil.cloud_node_mock(1)
arv_node = testutil.arvados_node_mock(1)
self.make_daemon([cloud_node], [arv_node])
- self.busywait(lambda: 1 == self.alive_monitor_count(),
- lambda: self.stop_proxy(self.daemon))
self.check_monitors_arvados_nodes(arv_node)
def test_node_pairing_after_arvados_update(self):
[testutil.arvados_node_mock(1, ip_address=None)])
arv_node = testutil.arvados_node_mock(2)
self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
- self.stop_proxy(self.daemon)
self.check_monitors_arvados_nodes(arv_node)
def test_arvados_node_un_and_re_paired(self):
self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
self.busywait(lambda: 0 == self.alive_monitor_count())
self.daemon.update_cloud_nodes([testutil.cloud_node_mock(3)])
- self.busywait(lambda: 1 == self.alive_monitor_count(),
- lambda: self.stop_proxy(self.daemon))
self.check_monitors_arvados_nodes(arv_node)
def test_old_arvados_node_not_double_assigned(self):
def test_node_count_satisfied(self):
self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1)],
want_sizes=[testutil.MockSize(1)])
- self.busywait(lambda: not self.node_setup.start.called,
- lambda: self.stop_proxy(self.daemon))
+ self.busywait(lambda: not self.node_setup.start.called)
def test_dont_count_missing_as_busy(self):
size = testutil.MockSize(1)
2,
last_ping_at='1970-01-01T01:02:03.04050607Z')],
want_sizes=[size, size])
- time.sleep(2)
+ self.busywait(lambda: 2 == self.alive_monitor_count())
self.busywait(lambda: self.node_setup.start.called)
def test_missing_counts_towards_max(self):
testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
want_sizes=[size, size],
max_nodes=2)
- self.busywait(lambda: not self.node_setup.start.called,
- lambda: self.stop_proxy(self.daemon))
+ self.busywait(lambda: not self.node_setup.start.called)
def test_excess_counts_missing(self):
size = testutil.MockSize(1)
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())
+ self.busywait(lambda: 2 == self.paired_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)
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.busywait(lambda: 2 == self.paired_monitor_count())
get_cloud_node = mock.MagicMock(name="get_cloud_node")
get_cloud_node.get.return_value = cloud_nodes[1]
mock_node_monitor = mock.MagicMock()
self.daemon.cloud_nodes.get()[cloud_nodes[1].id].shutdown_actor = mock_shutdown.proxy()
- self.assertEqual(2, self.alive_monitor_count())
+ self.busywait(lambda: 2 == self.alive_monitor_count())
for mon_ref in self.monitor_list():
self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
self.busywait(lambda: 1 == self.node_shutdown.start.call_count)
self.daemon.max_nodes.get(self.TIMEOUT)
self.assertTrue(self.node_setup.start.called)
self.daemon.update_server_wishlist(server_wishlist).get(self.TIMEOUT)
- self.busywait(lambda: 1 == self.node_setup.start.call_count,
- lambda: self.stop_proxy(self.daemon))
+ self.busywait(lambda: 1 == self.node_setup.start.call_count)
def test_boot_new_node_when_all_nodes_busy(self):
size = testutil.MockSize(2)
arv_node = testutil.arvados_node_mock(2, job_uuid=True)
self.make_daemon([testutil.cloud_node_mock(2, size=size)], [arv_node],
[size], avail_sizes=[(size, {"cores":1})])
- self.busywait(lambda: self.node_setup.start.called,
- lambda: self.stop_proxy(self.daemon))
+ self.busywait(lambda: 1 == self.paired_monitor_count())
+ self.busywait(lambda: self.node_setup.start.called)
def test_boot_new_node_below_min_nodes(self):
min_size = testutil.MockSize(1)
self.assertFalse(self.node_shutdown.start.called)
now = time.time()
self.monitor_list()[0].tell_proxy().consider_shutdown()
- self.busywait(lambda: self.node_shutdown.start.called,
- lambda: self.stop_proxy(self.daemon))
+ self.busywait(lambda: self.node_shutdown.start.called)
self.assertShutdownCancellable(False)
def test_booted_node_shut_down_when_never_paired(self):
self.assertEqual(1, self.alive_monitor_count())
self.daemon.update_cloud_nodes([cloud_node])
self.monitor_list()[0].tell_proxy().consider_shutdown()
- self.busywait(lambda: self.node_shutdown.start.called,
- lambda: self.stop_proxy(self.daemon))
+ self.busywait(lambda: self.node_shutdown.start.called)
self.assertShutdownCancellable(False)
def test_booted_node_shut_down_when_never_working(self):
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])
- self.busywait(lambda: self.node_shutdown.start.called,
- lambda: self.stop_proxy(self.daemon))
+ self.busywait(lambda: self.node_shutdown.start.called)
self.assertShutdownCancellable(False)
def test_node_that_pairs_not_considered_failed_boot(self):
def test_booting_nodes_shut_down(self):
self.make_daemon(want_sizes=[testutil.MockSize(1)])
self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
- self.busywait(lambda: self.last_setup.stop_if_no_cloud_node.called,
- lambda: self.stop_proxy(self.daemon))
+ self.busywait(lambda: self.last_setup.stop_if_no_cloud_node.called)
def test_all_booting_nodes_tried_to_shut_down(self):
size = testutil.MockSize(2)
arv_node = testutil.arvados_node_mock(1)
size = testutil.MockSize(1)
self.make_daemon(cloud_nodes=[cloud_node], arvados_nodes=[arv_node], want_sizes=[size])
- self.assertEqual(1, self.alive_monitor_count())
+ self.busywait(lambda: 1 == self.paired_monitor_count())
monitor = self.monitor_list()[0].proxy()
self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
self.stop_proxy(self.daemon)
cloud_node = testutil.cloud_node_mock(1)
arv_node = testutil.arvados_node_mock(1)
self.make_daemon(cloud_nodes=[cloud_node], arvados_nodes=[arv_node], min_nodes=1)
- self.assertEqual(1, self.alive_monitor_count())
+ self.busywait(lambda: 1 == self.paired_monitor_count())
monitor = self.monitor_list()[0].proxy()
self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
self.stop_proxy(self.daemon)
def test_shutdown_accepted_below_capacity(self):
self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
- self.assertEqual(1, self.alive_monitor_count())
+ self.busywait(lambda: 1 == self.alive_monitor_count())
monitor = self.monitor_list()[0].proxy()
self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
- self.busywait(lambda: self.node_shutdown.start.called,
- lambda: self.stop_proxy(self.daemon))
+ self.busywait(lambda: self.node_shutdown.start.called)
def test_shutdown_declined_when_idle_and_job_queued(self):
size = testutil.MockSize(1)
arv_nodes = [testutil.arvados_node_mock(3, job_uuid=True),
testutil.arvados_node_mock(4, job_uuid=None)]
self.make_daemon(cloud_nodes, arv_nodes, [size])
- self.daemon.ping().get(self.TIMEOUT)
- self.assertEqual(2, self.alive_monitor_count())
+ self.busywait(lambda: 2 == self.paired_monitor_count())
for mon_ref in self.monitor_list():
monitor = mon_ref.proxy()
if monitor.cloud_node.get(self.TIMEOUT) is cloud_nodes[-1]:
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.busywait(lambda: 1 == self.paired_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.busywait(lambda: 0 == self.alive_monitor_count())
+ self.busywait(lambda: 0 == self.paired_monitor_count())
def test_nodes_shutting_down_replaced_below_max_nodes(self):
size = testutil.MockSize(6)
self.assertTrue(self.node_shutdown.start.called)
self.daemon.update_server_wishlist(
[testutil.MockSize(6)]).get(self.TIMEOUT)
- self.busywait(lambda: self.node_setup.start.called,
- lambda: self.stop_proxy(self.daemon))
+ self.busywait(lambda: 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())
+ self.busywait(lambda: 1 == self.paired_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.busywait(lambda: not self.node_setup.start.called,
- lambda: self.stop_proxy(self.daemon))
+ self.busywait(lambda: not self.node_setup.start.called)
def test_nodes_shutting_down_count_against_excess(self):
size = testutil.MockSize(8)
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.busywait(lambda: 2 == self.alive_monitor_count())
+ self.busywait(lambda: 2 == self.paired_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)
size = testutil.MockSize(2)
self.daemon.update_server_wishlist([size] * 2).get(self.TIMEOUT)
self.timer.deliver()
- self.busywait(lambda: 1 == self.node_setup.start.call_count,
- lambda: self.stop_proxy(self.daemon))
+ self.busywait(lambda: 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()])
monitor = self.monitor_list()[0].proxy()
self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
- self.busywait(lambda: 1 == self.last_shutdown.stop.call_count,
- lambda: self.stop_proxy(self.daemon))
+ self.busywait(lambda: 1 == self.last_shutdown.stop.call_count)
def test_shutdown_actor_cleanup_copes_with_dead_actors(self):
self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
# the ActorDeadError.
self.last_shutdown.stop.side_effect = pykka.ActorDeadError
self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
- self.busywait(lambda: 1 == self.last_shutdown.stop.call_count,
- lambda: self.stop_proxy(self.daemon))
+ self.busywait(lambda: 1 == self.last_shutdown.stop.call_count)
def test_node_create_two_sizes(self):
small = testutil.MockSize(1)
testutil.arvados_node_mock(3)],
want_sizes=[small, small, big],
avail_sizes=avail_sizes)
-
+ self.busywait(lambda: 3 == self.paired_monitor_count())
self.daemon.update_server_wishlist([small, big, big]).get(self.TIMEOUT)
self.assertEqual(0, self.node_shutdown.start.call_count)
projects / arvados.git / blobdiff