remote_worker.py

   1 #!/usr/bin/env python3
   2
   3 """A simple utility to unpickle some code, run it, and pickle the
   4 results.
   5 """
   6
   7 import logging
   8 import os
   9 import platform
  10 import signal
  11 import threading
  12 import sys
  13 import time
  14
  15 import cloudpickle  # type: ignore
  16 import psutil  # type: ignore
  17
  18 import argparse_utils
  19 import bootstrap
  20 import config
  21 from thread_utils import background_thread
  22
  23
  24 logger = logging.getLogger(__file__)
  25
  26 cfg = config.add_commandline_args(
  27     f"Remote Worker ({__file__})",
  28     "Helper to run pickled code remotely and return results",
  29 )
  30 cfg.add_argument(
  31     '--code_file',
  32     type=str,
  33     required=True,
  34     metavar='FILENAME',
  35     help='The location of the bundle of code to execute.'
  36 )
  37 cfg.add_argument(
  38     '--result_file',
  39     type=str,
  40     required=True,
  41     metavar='FILENAME',
  42     help='The location where we should write the computation results.'
  43 )
  44 cfg.add_argument(
  45     '--watch_for_cancel',
  46     action=argparse_utils.ActionNoYes,
  47     default=False,
  48     help='Should we watch for the cancellation of our parent ssh process?'
  49 )
  50
  51
  52 @background_thread
  53 def watch_for_cancel(terminate_event: threading.Event) -> None:
  54     if platform.node() == 'VIDEO-COMPUTER':
  55         logger.warning('Background thread not allowed on retarded computers, sorry.')
  56         return
  57     logger.debug('Starting up background thread...')
  58     p = psutil.Process(os.getpid())
  59     while True:
  60         saw_sshd = False
  61         ancestors = p.parents()
  62         for ancestor in ancestors:
  63             name = ancestor.name()
  64             pid = ancestor.pid
  65             logger.debug(f'Ancestor process {name} (pid={pid})')
  66             if 'ssh' in name.lower():
  67                 saw_sshd = True
  68                 break
  69         if not saw_sshd:
  70             logger.error('Did not see sshd in our ancestors list?!  Committing suicide.')
  71             os.system('pstree')
  72             os.kill(os.getpid(), signal.SIGTERM)
  73             time.sleep(5.0)
  74             os.kill(os.getpid(), signal.SIGKILL)
  75             sys.exit(-1)
  76         if terminate_event.is_set():
  77             return
  78         time.sleep(1.0)
  79
  80
  81 @bootstrap.initialize
  82 def main() -> None:
  83     in_file = config.config['code_file']
  84     out_file = config.config['result_file']
  85
  86     (thread, stop_thread) = watch_for_cancel()
  87
  88     logger.debug(f'Reading {in_file}.')
  89     try:
  90         with open(in_file, 'rb') as rb:
  91             serialized = rb.read()
  92     except Exception as e:
  93         logger.exception(e)
  94         logger.critical(f'Problem reading {in_file}.  Aborting.')
  95         stop_thread.set()
  96         sys.exit(-1)
  97
  98     logger.debug(f'Deserializing {in_file}.')
  99     try:
 100         fun, args, kwargs = cloudpickle.loads(serialized)
 101     except Exception as e:
 102         logger.exception(e)
 103         logger.critical(f'Problem deserializing {in_file}.  Aborting.')
 104         stop_thread.set()
 105         sys.exit(-1)
 106
 107     logger.debug('Invoking user code...')
 108     start = time.time()
 109     ret = fun(*args, **kwargs)
 110     end = time.time()
 111     logger.debug(f'User code took {end - start:.1f}s')
 112
 113     logger.debug('Serializing results')
 114     try:
 115         serialized = cloudpickle.dumps(ret)
 116     except Exception as e:
 117         logger.exception(e)
 118         logger.critical(f'Could not serialize result ({type(ret)}).  Aborting.')
 119         stop_thread.set()
 120         sys.exit(-1)
 121
 122     logger.debug(f'Writing {out_file}.')
 123     try:
 124         with open(out_file, 'wb') as wb:
 125             wb.write(serialized)
 126     except Exception as e:
 127         logger.exception(e)
 128         logger.critical(f'Error writing {out_file}.  Aborting.')
 129         stop_thread.set()
 130         sys.exit(-1)
 131
 132     stop_thread.set()
 133     thread.join()
 134
 135
 136 if __name__ == '__main__':
 137     main()