3 """A simple utility to unpickle some code, run it, and pickle the
15 import cloudpickle # type: ignore
16 import psutil # type: ignore
21 from thread_utils import background_thread
24 logger = logging.getLogger(__file__)
26 cfg = config.add_commandline_args(
27 f"Remote Worker ({__file__})",
28 "Helper to run pickled code remotely and return results",
35 help='The location of the bundle of code to execute.'
42 help='The location where we should write the computation results.'
46 action=argparse_utils.ActionNoYes,
48 help='Should we watch for the cancellation of our parent ssh process?'
53 def watch_for_cancel(terminate_event: threading.Event) -> None:
54 logger.debug('Starting up background thread...')
55 p = psutil.Process(os.getpid())
58 ancestors = p.parents()
59 for ancestor in ancestors:
60 name = ancestor.name()
62 logger.debug(f'Ancestor process {name} (pid={pid})')
63 if 'ssh' in name.lower():
67 logger.error('Did not see sshd in our ancestors list?! Committing suicide.')
69 os.kill(os.getpid(), signal.SIGTERM)
71 os.kill(os.getpid(), signal.SIGKILL)
73 if terminate_event.is_set():
80 in_file = config.config['code_file']
81 out_file = config.config['result_file']
84 if config.config['watch_for_cancel']:
85 (thread, stop_thread) = watch_for_cancel()
87 logger.debug(f'Reading {in_file}.')
89 with open(in_file, 'rb') as rb:
90 serialized = rb.read()
91 except Exception as e:
93 logger.critical(f'Problem reading {in_file}. Aborting.')
97 logger.debug(f'Deserializing {in_file}.')
99 fun, args, kwargs = cloudpickle.loads(serialized)
100 except Exception as e:
102 logger.critical(f'Problem deserializing {in_file}. Aborting.')
106 logger.debug('Invoking user code...')
108 ret = fun(*args, **kwargs)
110 logger.debug(f'User code took {end - start:.1f}s')
112 logger.debug('Serializing results')
114 serialized = cloudpickle.dumps(ret)
115 except Exception as e:
117 logger.critical(f'Could not serialize result ({type(ret)}). Aborting.')
121 logger.debug(f'Writing {out_file}.')
123 with open(out_file, 'wb') as wb:
125 except Exception as e:
127 logger.critical(f'Error writing {out_file}. Aborting.')
131 if stop_thread is not None:
136 if __name__ == '__main__':