Used isort to sort imports. Also added to the git pre-commit hook.

[python_utils.git] / remote_worker.py

diff --git a/remote_worker.py b/remote_worker.py
index 84f8d56fa33318b507f3f11618c663861718182b..82b80ea3d722090ab7254eb24eac5884a9520172 100755 (executable)
--- a/remote_worker.py
+++ b/remote_worker.py
@@ -7,9 +7,10 @@ results.
 import logging
 import os
 import signal
-import threading
 import sys
+import threading
 import time
+from typing import Optional
 
 import cloudpickle  # type: ignore
 import psutil  # type: ignore
@@ -17,9 +18,9 @@ import psutil  # type: ignore
 import argparse_utils
 import bootstrap
 import config
+from stopwatch import Timer
 from thread_utils import background_thread
 
-
 logger = logging.getLogger(__file__)
 
 cfg = config.add_commandline_args(
@@ -31,35 +32,41 @@ cfg.add_argument(
     type=str,
     required=True,
     metavar='FILENAME',
-    help='The location of the bundle of code to execute.'
+    help='The location of the bundle of code to execute.',
 )
 cfg.add_argument(
     '--result_file',
     type=str,
     required=True,
     metavar='FILENAME',
-    help='The location where we should write the computation results.'
+    help='The location where we should write the computation results.',
 )
 cfg.add_argument(
     '--watch_for_cancel',
     action=argparse_utils.ActionNoYes,
-    default=False,
-    help='Should we watch for the cancellation of our parent ssh process?'
+    default=True,
+    help='Should we watch for the cancellation of our parent ssh process?',
 )
 
 
 @background_thread
 def watch_for_cancel(terminate_event: threading.Event) -> None:
+    logger.debug('Starting up background thread...')
     p = psutil.Process(os.getpid())
     while True:
         saw_sshd = False
         ancestors = p.parents()
         for ancestor in ancestors:
             name = ancestor.name()
+            pid = ancestor.pid
+            logger.debug(f'Ancestor process {name} (pid={pid})')
             if 'ssh' in name.lower():
                 saw_sshd = True
                 break
         if not saw_sshd:
+            logger.error(
+                'Did not see sshd in our ancestors list?!  Committing suicide.'
+            )
             os.system('pstree')
             os.kill(os.getpid(), signal.SIGTERM)
             time.sleep(5.0)
@@ -70,11 +77,28 @@ def watch_for_cancel(terminate_event: threading.Event) -> None:
         time.sleep(1.0)
 
 
+def cleanup_and_exit(
+    thread: Optional[threading.Thread],
+    stop_thread: Optional[threading.Event],
+    exit_code: int,
+) -> None:
+    if stop_thread is not None:
+        stop_thread.set()
+        assert thread is not None
+        thread.join()
+    sys.exit(exit_code)
+
+
 @bootstrap.initialize
 def main() -> None:
     in_file = config.config['code_file']
     out_file = config.config['result_file']
 
+    thread = None
+    stop_thread = None
+    if config.config['watch_for_cancel']:
+        (thread, stop_thread) = watch_for_cancel()
+
     logger.debug(f'Reading {in_file}.')
     try:
         with open(in_file, 'rb') as rb:
@@ -82,7 +106,7 @@ def main() -> None:
     except Exception as e:
         logger.exception(e)
         logger.critical(f'Problem reading {in_file}.  Aborting.')
-        sys.exit(-1)
+        cleanup_and_exit(thread, stop_thread, 1)
 
     logger.debug(f'Deserializing {in_file}.')
     try:
@@ -90,13 +114,12 @@ def main() -> None:
     except Exception as e:
         logger.exception(e)
         logger.critical(f'Problem deserializing {in_file}.  Aborting.')
-        sys.exit(-1)
+        cleanup_and_exit(thread, stop_thread, 2)
 
     logger.debug('Invoking user code...')
-    start = time.time()
-    ret = fun(*args, **kwargs)
-    end = time.time()
-    logger.debug(f'User code took {end - start:.1f}s')
+    with Timer() as t:
+        ret = fun(*args, **kwargs)
+    logger.debug(f'User code took {t():.1f}s')
 
     logger.debug('Serializing results')
     try:
@@ -104,7 +127,7 @@ def main() -> None:
     except Exception as e:
         logger.exception(e)
         logger.critical(f'Could not serialize result ({type(ret)}).  Aborting.')
-        sys.exit(-1)
+        cleanup_and_exit(thread, stop_thread, 3)
 
     logger.debug(f'Writing {out_file}.')
     try:
@@ -113,7 +136,8 @@ def main() -> None:
     except Exception as e:
         logger.exception(e)
         logger.critical(f'Error writing {out_file}.  Aborting.')
-        sys.exit(-1)
+        cleanup_and_exit(thread, stop_thread, 4)
+    cleanup_and_exit(thread, stop_thread, 0)
 
 
 if __name__ == '__main__':