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 signal
  10 import sys
  11 import threading
  12 import time
  13 from typing import Optional
  14
  15 import cloudpickle  # type: ignore
  16 import psutil  # type: ignore
  17
  18 import argparse_utils
  19 import bootstrap
  20 import config
  21 from stopwatch import Timer
  22 from thread_utils import background_thread
  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=True,
  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(
  68                 'Did not see sshd in our ancestors list?!  Committing suicide.'
  69             )
  70             os.system('pstree')
  71             os.kill(os.getpid(), signal.SIGTERM)
  72             time.sleep(5.0)
  73             os.kill(os.getpid(), signal.SIGKILL)
  74             sys.exit(-1)
  75         if terminate_event.is_set():
  76             return
  77         time.sleep(1.0)
  78
  79
  80 def cleanup_and_exit(
  81     thread: Optional[threading.Thread],
  82     stop_thread: Optional[threading.Event],
  83     exit_code: int,
  84 ) -> None:
  85     if stop_thread is not None:
  86         stop_thread.set()
  87         assert thread is not None
  88         thread.join()
  89     sys.exit(exit_code)
  90
  91
  92 @bootstrap.initialize
  93 def main() -> None:
  94     in_file = config.config['code_file']
  95     out_file = config.config['result_file']
  96
  97     thread = None
  98     stop_thread = None
  99     if config.config['watch_for_cancel']:
 100         (thread, stop_thread) = watch_for_cancel()
 101
 102     logger.debug(f'Reading {in_file}.')
 103     try:
 104         with open(in_file, 'rb') as rb:
 105             serialized = rb.read()
 106     except Exception as e:
 107         logger.exception(e)
 108         logger.critical(f'Problem reading {in_file}.  Aborting.')
 109         cleanup_and_exit(thread, stop_thread, 1)
 110
 111     logger.debug(f'Deserializing {in_file}.')
 112     try:
 113         fun, args, kwargs = cloudpickle.loads(serialized)
 114     except Exception as e:
 115         logger.exception(e)
 116         logger.critical(f'Problem deserializing {in_file}.  Aborting.')
 117         cleanup_and_exit(thread, stop_thread, 2)
 118
 119     logger.debug('Invoking user code...')
 120     with Timer() as t:
 121         ret = fun(*args, **kwargs)
 122     logger.debug(f'User code took {t():.1f}s')
 123
 124     logger.debug('Serializing results')
 125     try:
 126         serialized = cloudpickle.dumps(ret)
 127     except Exception as e:
 128         logger.exception(e)
 129         logger.critical(f'Could not serialize result ({type(ret)}).  Aborting.')
 130         cleanup_and_exit(thread, stop_thread, 3)
 131
 132     logger.debug(f'Writing {out_file}.')
 133     try:
 134         with open(out_file, 'wb') as wb:
 135             wb.write(serialized)
 136     except Exception as e:
 137         logger.exception(e)
 138         logger.critical(f'Error writing {out_file}.  Aborting.')
 139         cleanup_and_exit(thread, stop_thread, 4)
 140     cleanup_and_exit(thread, stop_thread, 0)
 141
 142
 143 if __name__ == '__main__':
 144     main()