services/nodemanager/arvnodeman/nodelist.py

   1 #!/usr/bin/env python
   2
   3 from __future__ import absolute_import, print_function
   4
   5 import subprocess
   6
   7 from . import clientactor
   8 from . import config
   9
  10 import arvados.util
  11
  12 class ArvadosNodeListMonitorActor(clientactor.RemotePollLoopActor):
  13     """Actor to poll the Arvados node list.
  14
  15     This actor regularly polls the list of Arvados node records, and
  16     sends it to subscribers.
  17     """
  18
  19     def is_common_error(self, exception):
  20         return isinstance(exception, config.ARVADOS_ERRORS)
  21
  22     def _item_key(self, node):
  23         return node['uuid']
  24
  25     def _send_request(self):
  26         nodelist = arvados.util.list_all(self._client.nodes().list)
  27
  28         # node hostname, state
  29         sinfo_out = subprocess.check_output(["sinfo", "--noheader", "--format=%n %t"])
  30         nodestates = {}
  31         for out in sinfo_out.splitlines():
  32             nodename, state = out.split(" ", 2)
  33             if state in ('alloc', 'comp'):
  34                 nodestates[nodename] = 'busy'
  35             elif state == 'idle':
  36                 nodestates[nodename] = 'idle'
  37             else:
  38                 nodestates[nodename] = 'down'
  39
  40         for n in nodelist:
  41             if n["slot_number"] and n["hostname"] and n["hostname"] in nodestates:
  42                 n["crunch_worker_state"] = nodestates[n["hostname"]]
  43             else:
  44                 n["crunch_worker_state"] = 'down'
  45
  46         return nodelist
  47
  48 class CloudNodeListMonitorActor(clientactor.RemotePollLoopActor):
  49     """Actor to poll the cloud node list.
  50
  51     This actor regularly polls the cloud to get a list of running compute
  52     nodes, and sends it to subscribers.
  53     """
  54
  55     def __init__(self, client, timer_actor, server_calc, *args, **kwargs):
  56         super(CloudNodeListMonitorActor, self).__init__(
  57             client, timer_actor, *args, **kwargs)
  58         self._calculator = server_calc
  59
  60     def is_common_error(self, exception):
  61         return self._client.is_cloud_exception(exception)
  62
  63     def _item_key(self, node):
  64         return node.id
  65
  66     def _send_request(self):
  67         nodes = self._client.list_nodes()
  68         for n in nodes:
  69             # Replace with libcloud NodeSize object with compatible
  70             # CloudSizeWrapper object which merges the size info reported from
  71             # the cloud with size information from the configuration file.
  72             n.size = self._calculator.find_size(n.size.id)
  73         return nodes