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     logger.debug('Starting up background thread...')
  55     p = psutil.Process(os.getpid())
  56     while True:
  57         saw_sshd = False
  58         ancestors = p.parents()
  59         for ancestor in ancestors:
  60             name = ancestor.name()
  61             pid = ancestor.pid
  62             logger.debug(f'Ancestor process {name} (pid={pid})')
  63             if 'ssh' in name.lower():
  64                 saw_sshd = True
  65                 break
  66         if not saw_sshd:
  67             logger.error('Did not see sshd in our ancestors list?!  Committing suicide.')
  68             os.system('pstree')
  69             os.kill(os.getpid(), signal.SIGTERM)
  70             time.sleep(5.0)
  71             os.kill(os.getpid(), signal.SIGKILL)
  72             sys.exit(-1)
  73         if terminate_event.is_set():
  74             return
  75         time.sleep(1.0)
  76
  77
  78 @bootstrap.initialize
  79 def main() -> None:
  80     in_file = config.config['code_file']
  81     out_file = config.config['result_file']
  82
  83     stop_thread = None
  84     if config.config['watch_for_cancel']:
  85         (thread, stop_thread) = watch_for_cancel()
  86
  87     logger.debug(f'Reading {in_file}.')
  88     try:
  89         with open(in_file, 'rb') as rb:
  90             serialized = rb.read()
  91     except Exception as e:
  92         logger.exception(e)
  93         logger.critical(f'Problem reading {in_file}.  Aborting.')
  94         stop_thread.set()
  95         sys.exit(-1)
  96
  97     logger.debug(f'Deserializing {in_file}.')
  98     try:
  99         fun, args, kwargs = cloudpickle.loads(serialized)
 100     except Exception as e:
 101         logger.exception(e)
 102         logger.critical(f'Problem deserializing {in_file}.  Aborting.')
 103         stop_thread.set()
 104         sys.exit(-1)
 105
 106     logger.debug('Invoking user code...')
 107     start = time.time()
 108     ret = fun(*args, **kwargs)
 109     end = time.time()
 110     logger.debug(f'User code took {end - start:.1f}s')
 111
 112     logger.debug('Serializing results')
 113     try:
 114         serialized = cloudpickle.dumps(ret)
 115     except Exception as e:
 116         logger.exception(e)
 117         logger.critical(f'Could not serialize result ({type(ret)}).  Aborting.')
 118         stop_thread.set()
 119         sys.exit(-1)
 120
 121     logger.debug(f'Writing {out_file}.')
 122     try:
 123         with open(out_file, 'wb') as wb:
 124             wb.write(serialized)
 125     except Exception as e:
 126         logger.exception(e)
 127         logger.critical(f'Error writing {out_file}.  Aborting.')
 128         stop_thread.set()
 129         sys.exit(-1)
 130
 131     if stop_thread is not None:
 132         stop_thread.set()
 133         thread.join()
 134
 135
 136 if __name__ == '__main__':
 137     main()