bpo-36533: Reinit logging.Handler locks on fork(). (GH-12704)

Instead of attempting to acquire and release them all across fork
which was leading to deadlocks in some applications that had chained
their own handlers while holding multiple locks.

Lib/logging/__init__.py

@@ -231,49 +231,38 @@ def _releaseLock():
 # Prevent a held logging lock from blocking a child from logging.
 
 if not hasattr(os, 'register_at_fork'):  # Windows and friends.
-    def _register_at_fork_acquire_release(instance):
+    def _register_at_fork_reinit_lock(instance):
         pass  # no-op when os.register_at_fork does not exist.
-else:  # The os.register_at_fork API exists
-    os.register_at_fork(before=_acquireLock,
-                        after_in_child=_releaseLock,
-                        after_in_parent=_releaseLock)
-
-    # A collection of instances with acquire and release methods (logging.Handler)
-    # to be called before and after fork.  The weakref avoids us keeping discarded
-    # Handler instances alive forever in case an odd program creates and destroys
-    # many over its lifetime.
-    _at_fork_acquire_release_weakset = weakref.WeakSet()
-
-
-    def _register_at_fork_acquire_release(instance):
-        # We put the instance itself in a single WeakSet as we MUST have only
-        # one atomic weak ref. used by both before and after atfork calls to
-        # guarantee matched pairs of acquire and release calls.
-        _at_fork_acquire_release_weakset.add(instance)
-
+else:
+    # A collection of instances with a createLock method (logging.Handler)
+    # to be called in the child after forking.  The weakref avoids us keeping
+    # discarded Handler instances alive.  A set is used to avoid accumulating
+    # duplicate registrations as createLock() is responsible for registering
+    # a new Handler instance with this set in the first place.
+    _at_fork_reinit_lock_weakset = weakref.WeakSet()
+
+    def _register_at_fork_reinit_lock(instance):
+        _acquireLock()
+        try:
+            _at_fork_reinit_lock_weakset.add(instance)
+        finally:
+            _releaseLock()
 
-    def _at_fork_weak_calls(method_name):
-        for instance in _at_fork_acquire_release_weakset:
-            method = getattr(instance, method_name)
+    def _after_at_fork_child_reinit_locks():
+        # _acquireLock() was called in the parent before forking.
+        for handler in _at_fork_reinit_lock_weakset:
             try:
-                method()
+                handler.createLock()
             except Exception as err:
                 # Similar to what PyErr_WriteUnraisable does.
                 print("Ignoring exception from logging atfork", instance,
-                      method_name, "method:", err, file=sys.stderr)
-
-
-    def _before_at_fork_weak_calls():
-        _at_fork_weak_calls('acquire')
+                      "._reinit_lock() method:", err, file=sys.stderr)
+        _releaseLock()  # Acquired by os.register_at_fork(before=.
 
 
-    def _after_at_fork_weak_calls():
-        _at_fork_weak_calls('release')
-
-
-    os.register_at_fork(before=_before_at_fork_weak_calls,
-                        after_in_child=_after_at_fork_weak_calls,
-                        after_in_parent=_after_at_fork_weak_calls)
+    os.register_at_fork(before=_acquireLock,
+                        after_in_child=_after_at_fork_child_reinit_locks,
+                        after_in_parent=_releaseLock)
 
 
 #---------------------------------------------------------------------------
@@ -900,7 +889,7 @@ def createLock(self):
         Acquire a thread lock for serializing access to the underlying I/O.
         """
         self.lock = threading.RLock()
-        _register_at_fork_acquire_release(self)
+        _register_at_fork_reinit_lock(self)
 
     def acquire(self):
         """

Lib/test/test_logging.py

@@ -668,10 +668,28 @@ def remove_loop(fname, tries):
     # register_at_fork mechanism is also present and used.
     @unittest.skipIf(not hasattr(os, 'fork'), 'Test requires os.fork().')
     def test_post_fork_child_no_deadlock(self):
-        """Ensure forked child logging locks are not held; bpo-6721."""
-        refed_h = logging.Handler()
+        """Ensure child logging locks are not held; bpo-6721 & bpo-36533."""
+        class _OurHandler(logging.Handler):
+            def __init__(self):
+                super().__init__()
+                self.sub_handler = logging.StreamHandler(
+                    stream=open('/dev/null', 'wt'))
+
+            def emit(self, record):
+                self.sub_handler.acquire()
+                try:
+                    self.sub_handler.emit(record)
+                finally:
+                    self.sub_handler.release()
+
+        self.assertEqual(len(logging._handlers), 0)
+        refed_h = _OurHandler()
         refed_h.name = 'because we need at least one for this test'
         self.assertGreater(len(logging._handlers), 0)
+        self.assertGreater(len(logging._at_fork_reinit_lock_weakset), 1)
+        test_logger = logging.getLogger('test_post_fork_child_no_deadlock')
+        test_logger.addHandler(refed_h)
+        test_logger.setLevel(logging.DEBUG)
 
         locks_held__ready_to_fork = threading.Event()
         fork_happened__release_locks_and_end_thread = threading.Event()
@@ -709,19 +727,24 @@ def lock_holder_thread_fn():
         locks_held__ready_to_fork.wait()
         pid = os.fork()
         if pid == 0:  # Child.
-            logging.error(r'Child process did not deadlock. \o/')
-            os._exit(0)
+            try:
+                test_logger.info(r'Child process did not deadlock. \o/')
+            finally:
+                os._exit(0)
         else:  # Parent.
+            test_logger.info(r'Parent process returned from fork. \o/')
             fork_happened__release_locks_and_end_thread.set()
             lock_holder_thread.join()
             start_time = time.monotonic()
             while True:
+                test_logger.debug('Waiting for child process.')
                 waited_pid, status = os.waitpid(pid, os.WNOHANG)
                 if waited_pid == pid:
                     break  # child process exited.
                 if time.monotonic() - start_time > 7:
                     break  # so long? implies child deadlock.
                 time.sleep(0.05)
+            test_logger.debug('Done waiting.')
             if waited_pid != pid:
                 os.kill(pid, signal.SIGKILL)
                 waited_pid, status = os.waitpid(pid, 0)

Misc/NEWS.d/next/Library/2019-04-06-00-55-09.bpo-36533.kzMyRH.rst

@@ -0,0 +1,6 @@
+Reinitialize logging.Handler locks in forked child processes instead of
+attempting to acquire them all in the parent before forking only to be
+released in the child process.  The acquire/release pattern was leading to
+deadlocks in code that has implemented any form of chained logging handlers
+that depend upon one another as the lock acquision order cannot be
+guaranteed.