Merge branch '13076-r-autogen-api'
[arvados.git] / services / nodemanager / tests / test_computenode_dispatch_slurm.py
index 73bcb577b7114d8812d2148f8257191e99231e9f..b61db5cba1c57918c622d1ee815461ba6fe6de77 100644 (file)
@@ -1,4 +1,7 @@
 #!/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
 
@@ -10,7 +13,10 @@ import mock
 
 import arvnodeman.computenode.dispatch.slurm as slurm_dispatch
 from . import testutil
-from .test_computenode_dispatch import ComputeNodeShutdownActorMixin, ComputeNodeUpdateActorTestCase
+from .test_computenode_dispatch import \
+    ComputeNodeShutdownActorMixin, \
+    ComputeNodeSetupActorTestCase, \
+    ComputeNodeUpdateActorTestCase
 
 @mock.patch('subprocess.check_output')
 class SLURMComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
@@ -23,15 +29,26 @@ class SLURMComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
         for s in args:
             self.assertIn(s, slurm_cmd)
 
-    def check_success_after_reset(self, proc_mock, end_state='drain\n'):
+    def check_success_after_reset(self, proc_mock, end_state='drain\n', timer=False):
         self.make_mocks(arvados_node=testutil.arvados_node_mock(63))
+        if not timer:
+            self.timer = testutil.MockTimer(False)
         self.make_actor()
         self.check_success_flag(None, 0)
+        # At this point, 1st try should have happened.
+
+        self.timer.deliver()
         self.check_success_flag(None, 0)
+        # At this point, 2nd try should have happened.
+
         # Order is critical here: if the mock gets called when no return value
         # or side effect is set, we may invoke a real subprocess.
         proc_mock.return_value = end_state
         proc_mock.side_effect = None
+
+        # 3rd try
+        self.timer.deliver()
+
         self.check_success_flag(True, 3)
         self.check_slurm_got_args(proc_mock, 'NodeName=compute63')
 
@@ -60,20 +77,18 @@ class SLURMComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
         self.check_success_flag(True)
         self.assertFalse(proc_mock.called)
 
-    def test_node_undrained_when_shutdown_cancelled(self, proc_mock):
+    def test_node_resumed_when_shutdown_cancelled(self, proc_mock):
         try:
             proc_mock.side_effect = iter(['', 'drng\n', 'drng\n', ''])
             self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
             self.timer = testutil.MockTimer(False)
             self.make_actor()
             self.busywait(lambda: proc_mock.call_args is not None)
-            self.shutdown_actor.cancel_shutdown("test").get(self.TIMEOUT)
+            self.shutdown_actor.cancel_shutdown("test")
             self.check_success_flag(False, 2)
-            self.assertEqual(proc_mock.call_args_list,
-                             [mock.call(['scontrol', 'update', 'NodeName=compute99', 'State=DRAIN', 'Reason=Node Manager shutdown']),
-                              mock.call(['sinfo', '--noheader', '-o', '%t', '-n', 'compute99']),
-                              mock.call(['sinfo', '--noheader', '-o', '%t', '-n', 'compute99']),
-                              mock.call(['scontrol', 'update', 'NodeName=compute99', 'State=RESUME'])])
+            self.assertEqual(proc_mock.call_args_list[0], mock.call(['scontrol', 'update', 'NodeName=compute99', 'State=DRAIN', 'Reason=Node Manager shutdown']))
+            self.assertEqual(proc_mock.call_args_list[-1], mock.call(['scontrol', 'update', 'NodeName=compute99', 'State=RESUME']))
+
         finally:
             self.shutdown_actor.actor_ref.stop()
 
@@ -81,11 +96,11 @@ class SLURMComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
         proc_mock.side_effect = iter([OSError, 'drain\n', OSError, 'idle\n', 'idle\n'])
         self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
         self.make_actor()
-        self.check_success_flag(False, 2)
+        self.check_success_flag(False, 5)
 
     def test_issue_slurm_drain_retry(self, proc_mock):
-        proc_mock.side_effect = iter([OSError, '', OSError, 'drng\n', 'drain\n', 'drain\n'])
-        self.check_success_after_reset(proc_mock)
+        proc_mock.side_effect = iter([OSError, OSError, 'drng\n', 'drain\n'])
+        self.check_success_after_reset(proc_mock, timer=False)
 
     def test_arvados_node_cleaned_after_shutdown(self, proc_mock):
         proc_mock.return_value = 'drain\n'
@@ -111,4 +126,18 @@ class SLURMComputeNodeUpdateActorTestCase(ComputeNodeUpdateActorTestCase):
         cloud_node = testutil.cloud_node_mock()
         arv_node = testutil.arvados_node_mock()
         self.updater.sync_node(cloud_node, arv_node).get(self.TIMEOUT)
-        check_output.assert_called_with(['scontrol', 'update', 'NodeName=compute99', 'Weight=99000'])
+        check_output.assert_called_with(['scontrol', 'update', 'NodeName=compute99', 'Weight=99000', 'Features=instancetype=z99.test'])
+
+class SLURMComputeNodeSetupActorTestCase(ComputeNodeSetupActorTestCase):
+    ACTOR_CLASS = slurm_dispatch.ComputeNodeSetupActor
+
+    @mock.patch('subprocess.check_output')
+    def test_update_node_features(self, check_output):
+        # `scontrol update` happens only if the Arvados node record
+        # has a hostname. ComputeNodeSetupActorTestCase.make_mocks
+        # uses mocks with scrubbed hostnames, so we override with the
+        # default testutil.arvados_node_mock.
+        self.make_mocks(arvados_effect=[testutil.arvados_node_mock()])
+        self.make_actor()
+        self.wait_for_assignment(self.setup_actor, 'cloud_node')
+        check_output.assert_called_with(['scontrol', 'update', 'NodeName=compute99', 'Weight=1000', 'Features=instancetype=z1.test'])