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             try:
  33                 nodename, state = out.split(" ", 2)
  34                 if state in ('alloc', 'alloc*',
  35                              'comp',  'comp*',
  36                              'mix',   'mix*',
  37                              'drng',  'drng*'):
  38                     nodestates[nodename] = 'busy'
  39                 elif state == 'idle':
  40                     nodestates[nodename] = 'idle'
  41                 else:
  42                     nodestates[nodename] = 'down'
  43             except ValueError:
  44                 pass
  45
  46         for n in nodelist:
  47             if n["slot_number"] and n["hostname"] and n["hostname"] in nodestates:
  48                 n["crunch_worker_state"] = nodestates[n["hostname"]]
  49             else:
  50                 n["crunch_worker_state"] = 'down'
  51
  52         return nodelist
  53
  54 class CloudNodeListMonitorActor(clientactor.RemotePollLoopActor):
  55     """Actor to poll the cloud node list.
  56
  57     This actor regularly polls the cloud to get a list of running compute
  58     nodes, and sends it to subscribers.
  59     """
  60
  61     def __init__(self, client, timer_actor, server_calc, *args, **kwargs):
  62         super(CloudNodeListMonitorActor, self).__init__(
  63             client, timer_actor, *args, **kwargs)
  64         self._calculator = server_calc
  65
  66     def is_common_error(self, exception):
  67         return self._client.is_cloud_exception(exception)
  68
  69     def _item_key(self, node):
  70         return node.id
  71
  72     def _send_request(self):
  73         nodes = self._client.list_nodes()
  74         for n in nodes:
  75             # Replace with libcloud NodeSize object with compatible
  76             # CloudSizeWrapper object which merges the size info reported from
  77             # the cloud with size information from the configuration file.
  78             n.size = self._calculator.find_size(n.size.id)
  79         return nodes