#!/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
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:
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
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
# 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"):