services/nodemanager/arvnodeman/jobqueue.py

   1 #!/usr/bin/env python
   2
   3 from __future__ import absolute_import, print_function
   4
   5 import logging
   6 import subprocess
   7
   8 from . import clientactor
   9 from .config import ARVADOS_ERRORS
  10
  11 class ServerCalculator(object):
  12     """Generate cloud server wishlists from an Arvados job queue.
  13
  14     Instantiate this class with a list of cloud node sizes you're willing to
  15     use, plus keyword overrides from the configuration.  Then you can pass
  16     job queues to servers_for_queue.  It will return a list of node sizes
  17     that would best satisfy the jobs, choosing the cheapest size that
  18     satisfies each job, and ignoring jobs that can't be satisfied.
  19     """
  20
  21     class CloudSizeWrapper(object):
  22         def __init__(self, real_size, **kwargs):
  23             self.real = real_size
  24             for name in ['id', 'name', 'ram', 'disk', 'bandwidth', 'price',
  25                          'extra']:
  26                 setattr(self, name, getattr(self.real, name))
  27             self.cores = kwargs.pop('cores')
  28             self.scratch = self.disk
  29             for name, override in kwargs.iteritems():
  30                 if not hasattr(self, name):
  31                     raise ValueError("unrecognized size field '%s'" % (name,))
  32                 setattr(self, name, override)
  33
  34             if self.price is None:
  35                 raise ValueError("Required field 'price' is None")
  36
  37         def meets_constraints(self, **kwargs):
  38             for name, want_value in kwargs.iteritems():
  39                 have_value = getattr(self, name)
  40                 if (have_value != 0) and (have_value < want_value):
  41                     return False
  42             return True
  43
  44
  45     def __init__(self, server_list, max_nodes=None, max_price=None):
  46         self.cloud_sizes = [self.CloudSizeWrapper(s, **kws)
  47                             for s, kws in server_list]
  48         self.cloud_sizes.sort(key=lambda s: s.price)
  49         self.max_nodes = max_nodes or float('inf')
  50         self.max_price = max_price or float('inf')
  51         self.logger = logging.getLogger('arvnodeman.jobqueue')
  52         self.logged_jobs = set()
  53
  54     @staticmethod
  55     def coerce_int(x, fallback):
  56         try:
  57             return int(x)
  58         except (TypeError, ValueError):
  59             return fallback
  60
  61     def cloud_size_for_constraints(self, constraints):
  62         want_value = lambda key: self.coerce_int(constraints.get(key), 0)
  63         wants = {'cores': want_value('min_cores_per_node'),
  64                  'ram': want_value('min_ram_mb_per_node'),
  65                  'scratch': want_value('min_scratch_mb_per_node')}
  66         for size in self.cloud_sizes:
  67             if size.meets_constraints(**wants):
  68                 return size
  69         return None
  70
  71     def servers_for_queue(self, queue):
  72         servers = []
  73         seen_jobs = set()
  74         for job in queue:
  75             seen_jobs.add(job['uuid'])
  76             constraints = job['runtime_constraints']
  77             want_count = max(1, self.coerce_int(constraints.get('min_nodes'), 1))
  78             cloud_size = self.cloud_size_for_constraints(constraints)
  79             if cloud_size is None:
  80                 if job['uuid'] not in self.logged_jobs:
  81                     self.logged_jobs.add(job['uuid'])
  82                     self.logger.debug("job %s not satisfiable", job['uuid'])
  83             elif (want_count <= self.max_nodes) and (want_count*cloud_size.price <= self.max_price):
  84                 servers.extend([cloud_size.real] * want_count)
  85         self.logged_jobs.intersection_update(seen_jobs)
  86         return servers
  87
  88     def cheapest_size(self):
  89         return self.cloud_sizes[0]
  90
  91     def find_size(self, sizeid):
  92         for s in self.cloud_sizes:
  93             if s.id == sizeid:
  94                 return s
  95         return None
  96
  97 class JobQueueMonitorActor(clientactor.RemotePollLoopActor):
  98     """Actor to generate server wishlists from the job queue.
  99
 100     This actor regularly polls Arvados' job queue, and uses the provided
 101     ServerCalculator to turn that into a list of requested node sizes.  That
 102     list is sent to subscribers on every poll.
 103     """
 104
 105     CLIENT_ERRORS = ARVADOS_ERRORS
 106
 107     def __init__(self, client, timer_actor, server_calc, *args, **kwargs):
 108         super(JobQueueMonitorActor, self).__init__(
 109             client, timer_actor, *args, **kwargs)
 110         self._calculator = server_calc
 111
 112     def _send_request(self):
 113         # cpus, memory, tempory disk space, reason, job name
 114         squeue_out = subprocess.check_output(["squeue", "--state=PENDING", "--noheader", "--format=%c %m %d %r %j"])
 115         queuelist = []
 116         for out in squeue_out.splitlines():
 117             cpu, ram, disk, reason, jobname = out.split(" ", 4)
 118             if reason in ("Resources", "ReqNodeNotAvail"):
 119                 queuelist.append({
 120                     "uuid": jobname,
 121                     "runtime_constraints": {
 122                         "min_cores_per_node": cpu,
 123                         "min_ram_mb_per_node": ram,
 124                         "min_scratch_mb_per_node": disk
 125                     }
 126                 })
 127
 128         queuelist.extend(self._client.jobs().queue().execute()['items'])
 129
 130         return queuelist
 131
 132     def _got_response(self, queue):
 133         server_list = self._calculator.servers_for_queue(queue)
 134         self._logger.debug("Calculated wishlist: %s",
 135                            ', '.join(s.name for s in server_list) or "(empty)")
 136         return super(JobQueueMonitorActor, self)._got_response(server_list)