services/nodemanager/arvnodeman/jobqueue.py

   1 #!/usr/bin/env python
   2
   3 from __future__ import absolute_import, print_function
   4
   5 import logging
   6
   7 from . import clientactor
   8 from .config import ARVADOS_ERRORS
   9
  10 class ServerCalculator(object):
  11     """Generate cloud server wishlists from an Arvados job queue.
  12
  13     Instantiate this class with a list of cloud node sizes you're willing to
  14     use, plus keyword overrides from the configuration.  Then you can pass
  15     job queues to servers_for_queue.  It will return a list of node sizes
  16     that would best satisfy the jobs, choosing the cheapest size that
  17     satisfies each job, and ignoring jobs that can't be satisfied.
  18     """
  19
  20     class CloudSizeWrapper(object):
  21         def __init__(self, real_size, **kwargs):
  22             self.real = real_size
  23             for name in ['id', 'name', 'ram', 'disk', 'bandwidth', 'price',
  24                          'extra']:
  25                 setattr(self, name, getattr(self.real, name))
  26             self.cores = kwargs.pop('cores')
  27             self.scratch = self.disk
  28             for name, override in kwargs.iteritems():
  29                 if not hasattr(self, name):
  30                     raise ValueError("unrecognized size field '%s'" % (name,))
  31                 setattr(self, name, override)
  32
  33         def meets_constraints(self, **kwargs):
  34             for name, want_value in kwargs.iteritems():
  35                 have_value = getattr(self, name)
  36                 if (have_value != 0) and (have_value < want_value):
  37                     return False
  38             return True
  39
  40
  41     def __init__(self, server_list, min_nodes=0, max_nodes=None):
  42         self.cloud_sizes = [self.CloudSizeWrapper(s, **kws)
  43                             for s, kws in server_list]
  44         self.cloud_sizes.sort(key=lambda s: s.price)
  45         self.min_nodes = min_nodes
  46         self.max_nodes = max_nodes or float('inf')
  47         self.logger = logging.getLogger('arvnodeman.jobqueue')
  48         self.logged_jobs = set()
  49
  50     @staticmethod
  51     def coerce_int(x, fallback):
  52         try:
  53             return int(x)
  54         except (TypeError, ValueError):
  55             return fallback
  56
  57     def cloud_size_for_constraints(self, constraints):
  58         want_value = lambda key: self.coerce_int(constraints.get(key), 0)
  59         wants = {'cores': want_value('min_cores_per_node'),
  60                  'ram': want_value('min_ram_mb_per_node'),
  61                  'scratch': want_value('min_scratch_mb_per_node')}
  62         for size in self.cloud_sizes:
  63             if size.meets_constraints(**wants):
  64                 return size
  65         return None
  66
  67     def servers_for_queue(self, queue):
  68         servers = []
  69         seen_jobs = set()
  70         for job in queue:
  71             seen_jobs.add(job['uuid'])
  72             constraints = job['runtime_constraints']
  73             want_count = self.coerce_int(constraints.get('min_nodes'), 1)
  74             cloud_size = self.cloud_size_for_constraints(constraints)
  75             if cloud_size is None:
  76                 if job['uuid'] not in self.logged_jobs:
  77                     self.logged_jobs.add(job['uuid'])
  78                     self.logger.debug("job %s not satisfiable", job['uuid'])
  79             elif (want_count <= self.max_nodes):
  80                 servers.extend([cloud_size.real] * max(1, want_count))
  81         self.logged_jobs.intersection_update(seen_jobs)
  82
  83         # Make sure the server queue has at least enough entries to
  84         # satisfy min_nodes.
  85         node_shortfall = self.min_nodes - len(servers)
  86         if node_shortfall > 0:
  87             basic_node = self.cloud_size_for_constraints({})
  88             servers.extend([basic_node.real] * node_shortfall)
  89         return servers
  90
  91
  92 class JobQueueMonitorActor(clientactor.RemotePollLoopActor):
  93     """Actor to generate server wishlists from the job queue.
  94
  95     This actor regularly polls Arvados' job queue, and uses the provided
  96     ServerCalculator to turn that into a list of requested node sizes.  That
  97     list is sent to subscribers on every poll.
  98     """
  99
 100     CLIENT_ERRORS = ARVADOS_ERRORS
 101     LOGGER_NAME = 'arvnodeman.jobqueue'
 102
 103     def __init__(self, client, timer_actor, server_calc, *args, **kwargs):
 104         super(JobQueueMonitorActor, self).__init__(
 105             client, timer_actor, *args, **kwargs)
 106         self._calculator = server_calc
 107
 108     def _send_request(self):
 109         return self._client.jobs().queue().execute()['items']
 110
 111     def _got_response(self, queue):
 112         server_list = self._calculator.servers_for_queue(queue)
 113         self._logger.debug("Sending server wishlist: %s",
 114                            ', '.join(s.name for s in server_list) or "(empty)")
 115         return super(JobQueueMonitorActor, self)._got_response(server_list)