Merge branch '12431-nm-delete-nic' refs #12431

[arvados.git] / services / nodemanager / arvnodeman / computenode / dispatch / slurm.py

diff --git a/services/nodemanager/arvnodeman/computenode/dispatch/slurm.py b/services/nodemanager/arvnodeman/computenode/dispatch/slurm.py
index 6c5cfc88b937a051ef14391325c1e6dfcb195896..c8883c3ae70f6614b7bd9063030c14e264dfe543 100644 (file)
--- a/services/nodemanager/arvnodeman/computenode/dispatch/slurm.py
+++ b/services/nodemanager/arvnodeman/computenode/dispatch/slurm.py
@@ -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
 
@@ -39,7 +42,7 @@ class ComputeNodeShutdownActor(SlurmMixin, ShutdownActorBase):
             self._logger.info("Draining SLURM node %s", self._nodename)
             self._later.issue_slurm_drain()
 
-    @RetryMixin._retry((subprocess.CalledProcessError,))
+    @RetryMixin._retry((subprocess.CalledProcessError, OSError))
     def cancel_shutdown(self, reason, try_resume=True):
         if self._nodename:
             if try_resume and self._get_slurm_state(self._nodename) in self.SLURM_DRAIN_STATES:
@@ -51,7 +54,7 @@ class ComputeNodeShutdownActor(SlurmMixin, ShutdownActorBase):
                 pass
         return super(ComputeNodeShutdownActor, self).cancel_shutdown(reason)
 
-    @RetryMixin._retry((subprocess.CalledProcessError,))
+    @RetryMixin._retry((subprocess.CalledProcessError, OSError))
     def issue_slurm_drain(self):
         if self.cancel_reason is not None:
             return
@@ -62,7 +65,7 @@ class ComputeNodeShutdownActor(SlurmMixin, ShutdownActorBase):
         else:
             self._later.shutdown_node()
 
-    @RetryMixin._retry((subprocess.CalledProcessError,))
+    @RetryMixin._retry((subprocess.CalledProcessError, OSError))
     def await_slurm_drain(self):
         if self.cancel_reason is not None:
             return
@@ -70,17 +73,24 @@ class ComputeNodeShutdownActor(SlurmMixin, ShutdownActorBase):
         if output in ("drng\n", "alloc\n", "drng*\n", "alloc*\n"):
             self._timer.schedule(time.time() + 10,
                                  self._later.await_slurm_drain)
-        elif output in ("idle\n"):
+        elif output in ("idle\n",):
             # Not in "drng" but idle, don't shut down
             self.cancel_shutdown("slurm state is %s" % output.strip(), try_resume=False)
         else:
             # any other state.
             self._later.shutdown_node()
 
+    def _destroy_node(self):
+        if self._nodename:
+            self._set_node_state(self._nodename, 'DOWN', 'Reason=Node Manager shutdown')
+        super(ComputeNodeShutdownActor, self)._destroy_node()
+
+
 class ComputeNodeUpdateActor(UpdateActorBase):
     def sync_node(self, cloud_node, arvados_node):
-        try:
-            subprocess.check_output(['scontrol', 'update', 'NodeName=' + arvados_node["hostname"], 'Weight=%i' % int(cloud_node.size.price * 1000)])
-        except OSError:
-            self._logger.warn("Unable to set slurm node weight.", exc_info=True)
+        if arvados_node.get("hostname"):
+            try:
+                subprocess.check_output(['scontrol', 'update', 'NodeName=' + arvados_node["hostname"], 'Weight=%i' % int(cloud_node.size.price * 1000)])
+            except:
+                self._logger.error("Unable to set slurm node weight.", exc_info=True)
         return super(ComputeNodeUpdateActor, self).sync_node(cloud_node, arvados_node)