[ARVADOS] created: 62991a76b32b9c7558dc359058d82a0be1da104d

[Git user]

        at  62991a76b32b9c7558dc359058d82a0be1da104d (commit)


commit 62991a76b32b9c7558dc359058d82a0be1da104d
Author: Brett Smith <brett at curoverse.com>
Date:   Tue Apr 12 18:22:59 2016 -0400

    8952: Node Manager shutdowns only check SLURM state of paired nodes.

diff --git a/services/nodemanager/arvnodeman/computenode/dispatch/slurm.py b/services/nodemanager/arvnodeman/computenode/dispatch/slurm.py
index 41919db..881a1c1 100644
--- a/services/nodemanager/arvnodeman/computenode/dispatch/slurm.py
+++ b/services/nodemanager/arvnodeman/computenode/dispatch/slurm.py
@@ -72,7 +72,7 @@ class ComputeNodeShutdownActor(SlurmMixin, ShutdownActorBase):
 class ComputeNodeMonitorActor(SlurmMixin, MonitorActorBase):
 
     def shutdown_eligible(self):
-        if self.arvados_node is not None:
+        if self.arvados_node and self.arvados_node['hostname']:
             state = self._get_slurm_state(self.arvados_node['hostname'])
             # Automatically eligible for shutdown if it's down or failed, but
             # not drain to avoid a race condition with resume_node().
diff --git a/services/nodemanager/tests/test_computenode_dispatch_slurm.py b/services/nodemanager/tests/test_computenode_dispatch_slurm.py
index 135b817..3859c4f 100644
--- a/services/nodemanager/tests/test_computenode_dispatch_slurm.py
+++ b/services/nodemanager/tests/test_computenode_dispatch_slurm.py
@@ -151,3 +151,27 @@ class SLURMComputeNodeMonitorActorTestCase(testutil.ActorTestMixin,
         check_output.return_value = "drain\n"
         self.make_actor(arv_node=testutil.arvados_node_mock())
         self.assertEquals('node is draining', self.node_actor.shutdown_eligible().get(self.TIMEOUT))
+
+    @mock.patch("subprocess.check_output")
+    def test_shutdown_node_that_never_pinged(self, check_output):
+        check_output.side_effect = Exception(
+            "tested code tried to call `sinfo` without a node name")
+        arv_node = testutil.arvados_node_mock()
+        arv_node.update(first_ping_at=None, last_ping_at=None, slot_number=None,
+                        hostname=None, ip_address=None)
+        self.make_actor(arv_node=arv_node)
+        self.shutdowns._set_state(True, 300)
+        self.assertIs(True, self.node_actor.shutdown_eligible().get(self.TIMEOUT))
+        check_output.assert_not_called()
+
+    @mock.patch("subprocess.check_output")
+    def test_no_shutdown_node_still_awaiting_ping(self, check_output):
+        check_output.side_effect = Exception(
+            "tested code tried to call `sinfo` without a node name")
+        arv_node = testutil.arvados_node_mock()
+        arv_node.update(first_ping_at=None, last_ping_at=None, slot_number=None,
+                        hostname=None, ip_address=None)
+        self.make_actor(arv_node=arv_node)
+        self.assertIn('shutdown window is not open',
+                      self.node_actor.shutdown_eligible().get(self.TIMEOUT))
+        check_output.assert_not_called()

-----------------------------------------------------------------------


hooks/post-receive
--