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_size=get_cloud_size,
actor_ref=mock_actor)
mock_actor.proxy.return_value = mock_proxy
+ mock_actor.tell_proxy.return_value = mock_proxy
self.last_setup = mock_proxy
return mock_actor
return mock_actor
def make_daemon(self, cloud_nodes=[], arvados_nodes=[], want_sizes=[],
- avail_sizes=[(testutil.MockSize(1), {"cores": 1})],
+ avail_sizes=None,
min_nodes=0, max_nodes=8,
shutdown_windows=[54, 5, 1],
max_total_price=None):
for name in ['cloud_nodes', 'arvados_nodes', 'server_wishlist']:
setattr(self, name + '_poller', mock.MagicMock(name=name + '_mock'))
+
+ if not avail_sizes:
+ if cloud_nodes or want_sizes:
+ avail_sizes=[(c.size, {"cores": int(c.id)}) for c in cloud_nodes] + [(s, {"cores": 1}) for s in want_sizes]
+ else:
+ avail_sizes=[(testutil.MockSize(1), {"cores": 1})]
+
self.arv_factory = mock.MagicMock(name='arvados_mock')
+ api_client = mock.MagicMock(name='api_client')
+ api_client.nodes().create().execute.side_effect = [testutil.arvados_node_mock(1),
+ testutil.arvados_node_mock(2)]
+ self.arv_factory.return_value = api_client
+
self.cloud_factory = mock.MagicMock(name='cloud_mock')
self.cloud_factory().node_start_time.return_value = time.time()
self.cloud_updates = mock.MagicMock(name='updates_mock')
self.timer = testutil.MockTimer(deliver_immediately=False)
+ self.cloud_factory().node_id.side_effect = lambda node: node.id
+ self.cloud_factory().broken.return_value = False
self.node_setup = mock.MagicMock(name='setup_mock')
self.node_setup.start.side_effect = self.mock_node_start
min_nodes, max_nodes, 600, 1800, 3600,
self.node_setup, self.node_shutdown,
max_total_price=max_total_price).proxy()
- if cloud_nodes is not None:
- self.daemon.update_cloud_nodes(cloud_nodes).get(self.TIMEOUT)
if arvados_nodes is not None:
self.daemon.update_arvados_nodes(arvados_nodes).get(self.TIMEOUT)
+ if cloud_nodes is not None:
+ self.daemon.update_cloud_nodes(cloud_nodes).get(self.TIMEOUT)
if want_sizes is not None:
self.daemon.update_server_wishlist(want_sizes).get(self.TIMEOUT)
def test_node_pairing_after_arvados_update(self):
cloud_node = testutil.cloud_node_mock(2)
self.make_daemon([cloud_node],
- [testutil.arvados_node_mock(2, ip_address=None)])
+ [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)
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)
+ self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(3)],
+ arvados_nodes=None)
arv_node = testutil.arvados_node_mock(3)
self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
self.check_monitors_arvados_nodes(arv_node)
def test_old_arvados_node_not_double_assigned(self):
arv_node = testutil.arvados_node_mock(3, age=9000)
size = testutil.MockSize(3)
- self.make_daemon(arvados_nodes=[arv_node], avail_sizes=[(size, {"cores":1})])
+ self.make_daemon(arvados_nodes=[arv_node],
+ avail_sizes=[(size, {"cores":1})])
self.daemon.update_server_wishlist([size]).get(self.TIMEOUT)
self.daemon.update_server_wishlist([size, size]).get(self.TIMEOUT)
self.stop_proxy(self.daemon)
self.assertIn(None, used_nodes)
def test_node_count_satisfied(self):
- self.make_daemon([testutil.cloud_node_mock()],
+ self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1)],
want_sizes=[testutil.MockSize(1)])
self.stop_proxy(self.daemon)
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)],
+ self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1, size=size),
+ testutil.cloud_node_mock(2, size=size)],
arvados_nodes=[testutil.arvados_node_mock(1),
- testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
+ 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)],
+ self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1, size=size),
+ testutil.cloud_node_mock(2, size=size)],
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],
def test_excess_counts_missing(self):
size = testutil.MockSize(1)
- cloud_nodes = [testutil.cloud_node_mock(1), testutil.cloud_node_mock(2)]
+ cloud_nodes = [testutil.cloud_node_mock(1, size=size), testutil.cloud_node_mock(2, size=size)]
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')],
def test_missing_shutdown_not_excess(self):
size = testutil.MockSize(1)
- cloud_nodes = [testutil.cloud_node_mock(1), testutil.cloud_node_mock(2)]
+ cloud_nodes = [testutil.cloud_node_mock(1, size=size), testutil.cloud_node_mock(2, size=size)]
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')],
mock_node_monitor.proxy.return_value = mock.NonCallableMock(cloud_node=get_cloud_node)
mock_shutdown = self.node_shutdown.start(node_monitor=mock_node_monitor)
- self.daemon.shutdowns.get()[cloud_nodes[1].id] = mock_shutdown.proxy()
- self.daemon.sizes_booting_shutdown.get()[cloud_nodes[1].id] = size
+ self.daemon.cloud_nodes.get()[cloud_nodes[1].id].shutdown_actor = mock_shutdown.proxy()
self.assertEqual(2, self.alive_monitor_count())
for mon_ref in self.monitor_list():
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)
self.stop_proxy(self.daemon)
self.assertTrue(self.node_setup.start.called)
for call in self.node_setup.start.call_args_list])
def test_no_new_node_when_ge_min_nodes_busy(self):
- cloud_nodes = [testutil.cloud_node_mock(n) for n in range(1, 4)]
+ size = testutil.MockSize(2)
+ cloud_nodes = [testutil.cloud_node_mock(n, size=size) for n in range(1, 4)]
arv_nodes = [testutil.arvados_node_mock(n, job_uuid=True)
for n in range(1, 4)]
self.make_daemon(cloud_nodes, arv_nodes, [], min_nodes=2)
self.assertEqual(0, self.node_setup.start.call_count)
def test_no_new_node_when_max_nodes_busy(self):
- self.make_daemon([testutil.cloud_node_mock(3)],
- [testutil.arvados_node_mock(3, job_uuid=True)],
- [testutil.MockSize(3)],
+ size = testutil.MockSize(3)
+ self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(3)],
+ arvados_nodes=[testutil.arvados_node_mock(3, job_uuid=True)],
+ want_sizes=[size],
max_nodes=1)
self.stop_proxy(self.daemon)
self.assertFalse(self.node_setup.start.called)
def start_node_boot(self, cloud_node=None, arv_node=None, id_num=1):
if cloud_node is None:
cloud_node = testutil.cloud_node_mock(id_num)
+ id_num = int(cloud_node.id)
if arv_node is None:
arv_node = testutil.arvados_node_mock(id_num)
self.make_daemon(want_sizes=[testutil.MockSize(id_num)],
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):
+ def test_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_cloud_nodes([cloud_node])
+ self.monitor_list()[0].proxy().cloud_node_start_time = time.time()-1801
self.daemon.update_server_wishlist(
- [testutil.MockSize(1)]).get(self.TIMEOUT)
+ [testutil.MockSize(4)]).get(self.TIMEOUT)
self.stop_proxy(self.daemon)
- self.assertEqual(1, self.node_setup.start.call_count)
+ self.assertEqual(2, 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
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))
+
def test_booted_node_lifecycle(self):
cloud_node = testutil.cloud_node_mock(6)
setup = self.start_node_boot(cloud_node, id_num=6)
shutdown = self.node_shutdown.start().proxy()
shutdown.cloud_node.get.return_value = cloud_node
self.daemon.node_finished_shutdown(shutdown).get(self.TIMEOUT)
+ self.daemon.update_cloud_nodes([])
self.assertTrue(shutdown.stop.called,
"shutdown actor not stopped after finishing")
self.assertTrue(monitor.actor_ref.actor_stopped.wait(self.TIMEOUT),
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.assertEqual(1, self.alive_monitor_count())
self.assertFalse(self.node_shutdown.start.called)
- self.timer.deliver()
+ now = time.time()
+ self.monitor_list()[0].tell_proxy().consider_shutdown()
+ self.busywait(lambda: self.node_shutdown.start.called)
self.stop_proxy(self.daemon)
self.assertShutdownCancellable(False)
def test_booted_node_shut_down_when_never_paired(self):
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.assertEqual(1, self.alive_monitor_count())
self.daemon.update_cloud_nodes([cloud_node])
- self.timer.deliver()
+ self.monitor_list()[0].tell_proxy().consider_shutdown()
+ self.busywait(lambda: self.node_shutdown.start.called)
self.stop_proxy(self.daemon)
self.assertShutdownCancellable(False)
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.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
self.daemon.node_up(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])
- self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
- self.timer.deliver()
+ self.busywait(lambda: self.node_shutdown.start.called)
self.stop_proxy(self.daemon)
self.assertShutdownCancellable(False)
def test_shutdown_declined_at_wishlist_capacity(self):
cloud_node = testutil.cloud_node_mock(1)
+ arv_node = testutil.arvados_node_mock(1)
size = testutil.MockSize(1)
- self.make_daemon(cloud_nodes=[cloud_node], want_sizes=[size])
+ self.make_daemon(cloud_nodes=[cloud_node], arvados_nodes=[arv_node], want_sizes=[size])
self.assertEqual(1, self.alive_monitor_count())
monitor = self.monitor_list()[0].proxy()
self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
def test_shutdown_declined_below_min_nodes(self):
cloud_node = testutil.cloud_node_mock(1)
- self.make_daemon(cloud_nodes=[cloud_node], min_nodes=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())
monitor = self.monitor_list()[0].proxy()
self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
self.assertTrue(self.node_shutdown.start.called)
def test_shutdown_declined_when_idle_and_job_queued(self):
- cloud_nodes = [testutil.cloud_node_mock(n) for n in [3, 4]]
+ size = testutil.MockSize(1)
+ cloud_nodes = [testutil.cloud_node_mock(n, size=size) for n in [3, 4]]
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, [testutil.MockSize(1)])
+ self.make_daemon(cloud_nodes, arv_nodes, [size])
self.assertEqual(2, self.alive_monitor_count())
for mon_ref in self.monitor_list():
monitor = mon_ref.proxy()
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)],
+ self.make_daemon([cloud_node], [testutil.arvados_node_mock(6, crunch_worker_state='down')],
avail_sizes=[(size, {"cores":1})])
self.assertEqual(1, self.alive_monitor_count())
monitor = self.monitor_list()[0].proxy()
self.stop_proxy(self.daemon)
self.assertEqual(1, self.last_shutdown.stop.call_count)
- def busywait(self, f):
- n = 0
- while not f() and n < 10:
- time.sleep(.1)
- n += 1
- self.assertTrue(f())
-
def test_node_create_two_sizes(self):
small = testutil.MockSize(1)
big = testutil.MockSize(2)
self.daemon.node_can_shutdown(c.actor)
booting = self.daemon.booting.get()
- shutdowns = self.daemon.shutdowns.get()
+ cloud_nodes = self.daemon.cloud_nodes.get()
self.stop_proxy(self.daemon)
# shutting down a small node
sizecounts = {a[0].id: 0 for a in avail_sizes}
- for b in shutdowns.itervalues():
- sizecounts[b.cloud_node.get().size.id] += 1
+ for b in cloud_nodes.nodes.itervalues():
+ if b.shutdown_actor is not None:
+ sizecounts[b.cloud_node.size.id] += 1
self.assertEqual(1, sizecounts[small.id])
self.assertEqual(0, sizecounts[big.id])
projects / arvados.git / blobdiff