bpo-31308: If multiprocessing's forkserver dies, launch it again when necessary (#3246)

* bpo-31308: If multiprocessing's forkserver dies, launch it again when necessary.

* Fix test on Windows

* Add NEWS entry

* Adopt a different approach: ignore SIGINT and SIGTERM, as in semaphore tracker.

* Fix comment

* Make sure the test doesn't muck with process state

* Also test previously-started processes

* Update 2017-08-30-17-59-36.bpo-31308.KbexyC.rst

* Avoid masking SIGTERM in forkserver.  It's not necessary and causes a race condition in test_many_processes.
This commit is contained in:
Antoine Pitrou 2017-11-03 13:34:22 +01:00 committed by GitHub
parent 4f57409a2f
commit fc6b348b12
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 66 additions and 5 deletions

View File

@ -34,6 +34,7 @@ class ForkServer(object):
def __init__(self):
self._forkserver_address = None
self._forkserver_alive_fd = None
self._forkserver_pid = None
self._inherited_fds = None
self._lock = threading.Lock()
self._preload_modules = ['__main__']
@ -90,8 +91,17 @@ class ForkServer(object):
'''
with self._lock:
semaphore_tracker.ensure_running()
if self._forkserver_alive_fd is not None:
return
if self._forkserver_pid is not None:
# forkserver was launched before, is it still running?
pid, status = os.waitpid(self._forkserver_pid, os.WNOHANG)
if not pid:
# still alive
return
# dead, launch it again
os.close(self._forkserver_alive_fd)
self._forkserver_address = None
self._forkserver_alive_fd = None
self._forkserver_pid = None
cmd = ('from multiprocessing.forkserver import main; ' +
'main(%d, %d, %r, **%r)')
@ -127,6 +137,7 @@ class ForkServer(object):
os.close(alive_r)
self._forkserver_address = address
self._forkserver_alive_fd = alive_w
self._forkserver_pid = pid
#
#
@ -157,11 +168,11 @@ def main(listener_fd, alive_r, preload, main_path=None, sys_path=None):
# Dummy signal handler, doesn't do anything
pass
# letting SIGINT through avoids KeyboardInterrupt tracebacks
# unblocking SIGCHLD allows the wakeup fd to notify our event loop
handlers = {
# unblocking SIGCHLD allows the wakeup fd to notify our event loop
signal.SIGCHLD: sigchld_handler,
signal.SIGINT: signal.SIG_DFL,
# protect the process from ^C
signal.SIGINT: signal.SIG_IGN,
}
old_handlers = {sig: signal.signal(sig, val)
for (sig, val) in handlers.items()}

View File

@ -603,6 +603,54 @@ class _TestProcess(BaseTestCase):
finally:
setattr(sys, stream_name, old_stream)
@classmethod
def _sleep_and_set_event(self, evt, delay=0.0):
time.sleep(delay)
evt.set()
def check_forkserver_death(self, signum):
# bpo-31308: if the forkserver process has died, we should still
# be able to create and run new Process instances (the forkserver
# is implicitly restarted).
if self.TYPE == 'threads':
self.skipTest('test not appropriate for {}'.format(self.TYPE))
sm = multiprocessing.get_start_method()
if sm != 'forkserver':
# The fork method by design inherits all fds from the parent,
# trying to go against it is a lost battle
self.skipTest('test not appropriate for {}'.format(sm))
from multiprocessing.forkserver import _forkserver
_forkserver.ensure_running()
evt = self.Event()
proc = self.Process(target=self._sleep_and_set_event, args=(evt, 1.0))
proc.start()
pid = _forkserver._forkserver_pid
os.kill(pid, signum)
time.sleep(1.0) # give it time to die
evt2 = self.Event()
proc2 = self.Process(target=self._sleep_and_set_event, args=(evt2,))
proc2.start()
proc2.join()
self.assertTrue(evt2.is_set())
self.assertEqual(proc2.exitcode, 0)
proc.join()
self.assertTrue(evt.is_set())
self.assertIn(proc.exitcode, (0, 255))
def test_forkserver_sigint(self):
# Catchable signal
self.check_forkserver_death(signal.SIGINT)
def test_forkserver_sigkill(self):
# Uncatchable signal
if os.name != 'nt':
self.check_forkserver_death(signal.SIGKILL)
#
#

View File

@ -0,0 +1,2 @@
Make multiprocessing's forkserver process immune to Ctrl-C and other user interruptions.
If it crashes, restart it when necessary.