Merge branch '8784-dir-listings'

[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 cbeabd1a8ad7697fcc0783647f46aee13222bc23..fa56578cffa1108526584ded9730a9cb5ffbbda9 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
 
@@ -6,8 +9,9 @@ import subprocess
 import time
 
 from . import \
-    ComputeNodeSetupActor, ComputeNodeUpdateActor, ComputeNodeMonitorActor
+    ComputeNodeSetupActor, ComputeNodeMonitorActor
 from . import ComputeNodeShutdownActor as ShutdownActorBase
+from . import ComputeNodeUpdateActor as UpdateActorBase
 from .. import RetryMixin
 
 class SlurmMixin(object):
@@ -38,7 +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:
@@ -50,7 +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
@@ -61,7 +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
@@ -75,3 +79,18 @@ class ComputeNodeShutdownActor(SlurmMixin, ShutdownActorBase):
         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):
+        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)