3 """A simple utility to unpickle some code, run it, and pickle the
13 from typing import Optional
15 import cloudpickle # type: ignore
16 import psutil # type: ignore
21 from stopwatch import Timer
22 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('Ancestor process %s (pid=%d)', name, 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():
79 thread: Optional[threading.Thread],
80 stop_thread: Optional[threading.Event],
83 if stop_thread is not None:
85 assert thread is not None
92 in_file = config.config['code_file']
93 out_file = config.config['result_file']
97 if config.config['watch_for_cancel']:
98 (thread, stop_thread) = watch_for_cancel()
100 logger.debug('Reading %s.', in_file)
102 with open(in_file, 'rb') as rb:
103 serialized = rb.read()
104 except Exception as e:
106 logger.critical('Problem reading %s. Aborting.', in_file)
107 cleanup_and_exit(thread, stop_thread, 1)
109 logger.debug('Deserializing %s', in_file)
111 fun, args, kwargs = cloudpickle.loads(serialized)
112 except Exception as e:
114 logger.critical('Problem deserializing %s. Aborting.', in_file)
115 cleanup_and_exit(thread, stop_thread, 2)
117 logger.debug('Invoking user code...')
119 ret = fun(*args, **kwargs)
120 logger.debug('User code took %.1fs', t())
122 logger.debug('Serializing results')
124 serialized = cloudpickle.dumps(ret)
125 except Exception as e:
127 logger.critical('Could not serialize result (%s). Aborting.', type(ret))
128 cleanup_and_exit(thread, stop_thread, 3)
130 logger.debug('Writing %s', out_file)
132 with open(out_file, 'wb') as wb:
134 except Exception as e:
136 logger.critical('Error writing %s. Aborting.', out_file)
137 cleanup_and_exit(thread, stop_thread, 4)
138 cleanup_and_exit(thread, stop_thread, 0)
141 if __name__ == '__main__':