4380: Add SLURM dispatcher to Node Manager.
[arvados.git] / services / nodemanager / arvnodeman / computenode / dispatch / slurm.py
1 #!/usr/bin/env python
2
3 from __future__ import absolute_import, print_function
4
5 import subprocess
6 import time
7
8 from . import \
9     ComputeNodeSetupActor, ComputeNodeUpdateActor, ComputeNodeMonitorActor
10 from . import ComputeNodeShutdownActor as ShutdownActorBase
11
12 class ComputeNodeShutdownActor(ShutdownActorBase):
13     def on_start(self):
14         arv_node = self._monitor.arvados_node.get()
15         if arv_node is None:
16             return super(ComputeNodeShutdownActor, self).on_start()
17         else:
18             self._nodename = arv_node['hostname']
19             self._logger.info("Draining SLURM node %s", self._nodename)
20             self._later.issue_slurm_drain()
21
22     def _set_node_state(self, state, *args):
23         cmd = ['scontrol', 'update', 'NodeName=' + self._nodename,
24                'State=' + state]
25         cmd.extend(args)
26         subprocess.check_output(cmd)
27
28     @ShutdownActorBase._retry((subprocess.CalledProcessError,))
29     def cancel_shutdown(self):
30         self._set_node_state('RESUME')
31         return super(ComputeNodeShutdownActor, self).cancel_shutdown()
32
33     @ShutdownActorBase._stop_if_window_closed
34     @ShutdownActorBase._retry((subprocess.CalledProcessError,))
35     def issue_slurm_drain(self):
36         self._set_node_state('DRAIN', 'Reason=Node Manager shutdown')
37         self._logger.info("Waiting for SLURM node %s to drain", self._nodename)
38         self._later.await_slurm_drain()
39
40     @ShutdownActorBase._stop_if_window_closed
41     @ShutdownActorBase._retry((subprocess.CalledProcessError,))
42     def await_slurm_drain(self):
43         output = subprocess.check_output(
44             ['sinfo', '--noheader', '-o', '%t', '-n', self._nodename])
45         if output == 'drain\n':
46             self._later.shutdown_node()
47         else:
48             self._timer.schedule(time.time() + 10,
49                                  self._later.await_slurm_drain)