Not wait cloudtail finish forerver on Windows.

scripts/slave/recipe_modules/goma/resources/cloudtail_utils.py

 #!/usr/bin/env python
 # Copyright (c) 2016 The Chromium Authors. All rights reserved.
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.
 
 import argparse
 import errno
+import multiprocessing
 import os
 import signal
 import subprocess
 import sys
 import time
 
 from slave import goma_utils
 
 
 def start_cloudtail(args):
@@ skipping 29 matching lines @@
       return False
     raise e
   return True
 
 
 class NotDiedError(Exception):
   def __str__(self):
     return "NotDiedError"
 
 
+def wait_termination_win(pid):
+  """Send CTRL_C_EVENT or SIGINT to pid and wait termination of pid.
+
+  Args:
+    pid(int): pid of process which this function waits termination.
+
+  Returns:
+    None: if pid has already finished or process finished succesfully.
+    instance of Exception: if raised unexpectedly.
+  """
+  import win32api
+  import win32con
+  import pywintypes
+  handle = None
+  try:
+    handle = win32api.OpenProcess(
+        win32con.PROCESS_QUERY_INFORMATION | win32con.SYNCHRONIZE,
+        False, pid)
+    try:
+      os.kill(pid, signal.CTRL_C_EVENT)
+      print('CTRL_C_EVENT has been sent to process %d. '
+            'Going to wait for the process finishes.' % pid)
+    except WindowsError as e:  # pylint: disable=E0602
+      # If a target process does not share terminal, we cannot send Ctrl-C.
+      if e.errno == 87:  # 87 == invalid parameter
+        os.kill(pid, signal.SIGINT)
+    os.waitpid(handle, 0)

[Vadim Sh., 2016/10/25 18:35:41]

I think it would be simpler to use WaitForSingleObject with timeout instead of
'multiprocessing' (now that use have process HANDLE and access to WinAPI).

[Yoshisato Yanagisawa, 2016/10/26 01:17:45]

Done.

+    return None
+  except pywintypes.error as e:
+    if e[0] == 87 and e[1] == 'OpenProcess':
+      print('Can\'t open process %d. Already dead? error %d.' % (pid, e))
+      return None
+    raise
+  except OSError as e:
+    if e.errno in (errno.ECHILD, errno.EPERM, errno.ESRCH):
+      print('Can\'t send SIGINT to process %d. Already dead? Errno %d.' %
+            (pid, e.errno))
+      return None
+    raise
+  except Exception as e:
+    return e

[Vadim Sh., 2016/10/25 18:35:41]

this is weird... is it because of 'multiprocessing'?

[Yoshisato Yanagisawa, 2016/10/26 01:17:45]

right.  revised without multiprocessing.

+  finally:
+    if handle:
+      win32api.CloseHandle(handle)
+
+
 def wait_termination(pid):
   """Send SIGINT to pid and wait termination of pid.
 
   Args:
     pid(int): pid of process which this function waits termination.
 
   Raises:
     OSError: is_running_posix, os.waitpid and os.kill may throw OSError.
     NotDiedError: if cloudtail is running after 10 seconds waiting,
                   NotDiedError is raised.
   """
-
-  try:
-    os.kill(pid, signal.SIGINT)
-  except OSError as e:
-    # Already dead?
-    if e.errno in (errno.ECHILD, errno.EPERM, errno.ESRCH):
-      print('Can\'t send SIGINT to process %d. Already dead? Errno %d.' %
-            (pid, e.errno))
+  if os.name == 'nt':
+    pool = multiprocessing.Pool(1)
+    res = pool.apply_async(wait_termination_win, [pid])
+    try:
+      e = res.get(10)
+    except multiprocessing.TimeoutError:
+      print('process %d running more than 10 seconds.' % pid)
+      raise NotDiedError()
+    if e is None:
       return
-    raise
-
-  print('SIGINT has been sent to process %d. '
-        'Going to wait for the process finishes.' % pid)
-  if os.name == 'nt':
+    raise e
+  else:
     try:
-      os.waitpid(pid, 0)
+      os.kill(pid, signal.SIGINT)
     except OSError as e:
-      if e.errno == errno.ECHILD:
-        print('process %d died before waitpitd' % pid)
+      if e.errno in (errno.ECHILD, errno.EPERM, errno.ESRCH):
+        print('Can\'t send SIGINT to process %d. Already dead? Errno %d.' %
+              (pid, e.errno))
         return
-      raise e
-  else:
     for _ in xrange(10):
       time.sleep(1)
       if not is_running_posix(pid):
         return
-
     print('process %d running more than 10 seconds' % pid)
     raise NotDiedError()
 
 
 def main():
   parser = argparse.ArgumentParser(
       description='cloudtail utility for goma recipe module.')
 
   subparsers = parser.add_subparsers(help='commands for cloudtail')
 
@@ skipping 15 matching lines @@
 
   if args.command == 'start':
     start_cloudtail(args)
   elif args.command == 'stop':
     with open(args.killed_pid_file) as f:
       # cloudtail flushes log and terminates
       # within 5 seconds when it recieves SIGINT.
       pid = int(f.read())
     try:
       wait_termination(pid)
-    except (OSError, NotDiedError) as e:
+    except Exception as e:
       print('Going to send SIGTERM to process %d due to Error %s' % (pid, e))
       # Since Windows does not have SIGKILL, we need to use SIGTERM.
       try:
         os.kill(pid, signal.SIGTERM)
       except OSError as e:
         print('Failed to send SIGTERM to process %d: %s' % (pid, e))
       # We do not reraise because I believe not suspending the process
       # is more important than completely killing cloudtail.
 
 
 if '__main__' == __name__:
   sys.exit(main())