Support gRPC Python client-side fork with epoll1

A process may fork after invoking grpc_init() and use gRPC in the child
if and only if the child process first destroys all gRPC resources
inherited from the parent process and invokes grpc_shutdown().
Subsequent to this, the child will be able to re-initialize and use
gRPC. After fork, the parent process will be able to continue to use
existing gRPC resources such as channels and calls without interference
from the child process.

To facilitate gRPC Python applications meeting the above constraints,
gRPC Python will automatically destroy and shutdown all gRPC Core
resources in the child's post-fork handler, including cancelling
in-flight calls (see detailed design below). From the client's
perspective, the child process is now free to create new channels and
use gRPC.

1 files changed, 203 insertions, 0 deletions

diff --git a/src/python/grpcio/grpc/_cython/_cygrpc/fork_posix.pyx.pxi b/src/python/grpcio/grpc/_cython/_cygrpc/fork_posix.pyx.pxi
new file mode 100644
index 0000000000..1176258da8
--- /dev/null
+++ b/src/python/grpcio/grpc/_cython/_cygrpc/fork_posix.pyx.pxi
@@ -0,0 +1,203 @@
+# Copyright 2018 gRPC authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import logging
+import os
+import threading
+
+_LOGGER = logging.getLogger(__name__)
+
+_AWAIT_THREADS_TIMEOUT_SECONDS = 5
+
+_TRUE_VALUES = ['yes',  'Yes',  'YES', 'true', 'True', 'TRUE', '1']
+
+# This flag enables experimental support within gRPC Python for applications
+# that will fork() without exec(). When enabled, gRPC Python will attempt to
+# pause all of its internally created threads before the fork syscall proceeds.
+#
+# For this to be successful, the application must not have multiple threads of
+# its own calling into gRPC when fork is invoked. Any callbacks from gRPC
+# Python-spawned threads into user code (e.g., callbacks for asynchronous RPCs)
+# must  not block and should execute quickly.
+#
+# This flag is not supported on Windows.
+_GRPC_ENABLE_FORK_SUPPORT = (
+    os.environ.get('GRPC_ENABLE_FORK_SUPPORT', '0')
+        .lower() in _TRUE_VALUES)
+
+_GRPC_POLL_STRATEGY = os.environ.get('GRPC_POLL_STRATEGY')
+
+cdef void __prefork() nogil:
+    with gil:
+        with _fork_state.fork_in_progress_condition:
+            _fork_state.fork_in_progress = True
+        if not _fork_state.active_thread_count.await_zero_threads(
+                _AWAIT_THREADS_TIMEOUT_SECONDS):
+            _LOGGER.error(
+                'Failed to shutdown gRPC Python threads prior to fork. '
+                'Behavior after fork will be undefined.')
+
+
+cdef void __postfork_parent() nogil:
+    with gil:
+        with _fork_state.fork_in_progress_condition:
+            _fork_state.fork_in_progress = False
+            _fork_state.fork_in_progress_condition.notify_all()
+
+
+cdef void __postfork_child() nogil:
+    with gil:
+        # Thread could be holding the fork_in_progress_condition inside of
+        # block_if_fork_in_progress() when fork occurs. Reset the lock here.
+        _fork_state.fork_in_progress_condition = threading.Condition()
+        # A thread in return_from_user_request_generator() may hold this lock
+        # when fork occurs.
+        _fork_state.active_thread_count = _ActiveThreadCount()
+        for state_to_reset in _fork_state.postfork_states_to_reset:
+            state_to_reset.reset_postfork_child()
+        _fork_state.fork_epoch += 1
+        for channel in _fork_state.channels:
+            channel._close_on_fork()
+        # TODO(ericgribkoff) Check and abort if core is not shutdown
+        with _fork_state.fork_in_progress_condition:
+            _fork_state.fork_in_progress = False
+
+
+def fork_handlers_and_grpc_init():
+    grpc_init()
+    if _GRPC_ENABLE_FORK_SUPPORT:
+        # TODO(ericgribkoff) epoll1 is default for grpcio distribution. Decide whether to expose
+        # grpc_get_poll_strategy_name() from ev_posix.cc to get actual polling choice.
+        if _GRPC_POLL_STRATEGY is not None and _GRPC_POLL_STRATEGY != "epoll1":
+            _LOGGER.error(
+                'gRPC Python fork support is only compatible with the epoll1 '
+                'polling engine')
+            return
+        with _fork_state.fork_handler_registered_lock:
+            if not _fork_state.fork_handler_registered:
+                pthread_atfork(&__prefork, &__postfork_parent, &__postfork_child)
+                _fork_state.fork_handler_registered = True
+
+
+class ForkManagedThread(object):
+    def __init__(self, target, args=()):
+        if _GRPC_ENABLE_FORK_SUPPORT:
+            def managed_target(*args):
+                try:
+                    target(*args)
+                finally:
+                    _fork_state.active_thread_count.decrement()
+            self._thread = threading.Thread(target=managed_target, args=args)
+        else:
+            self._thread = threading.Thread(target=target, args=args)
+
+    def setDaemon(self, daemonic):
+        self._thread.daemon = daemonic
+
+    def start(self):
+        if _GRPC_ENABLE_FORK_SUPPORT:
+            _fork_state.active_thread_count.increment()
+        self._thread.start()
+
+    def join(self):
+        self._thread.join()
+
+
+def block_if_fork_in_progress(postfork_state_to_reset=None):
+    if _GRPC_ENABLE_FORK_SUPPORT:
+        with _fork_state.fork_in_progress_condition:
+            if not _fork_state.fork_in_progress:
+                return
+            if postfork_state_to_reset is not None:
+                _fork_state.postfork_states_to_reset.append(postfork_state_to_reset)
+            _fork_state.active_thread_count.decrement()
+            _fork_state.fork_in_progress_condition.wait()
+            _fork_state.active_thread_count.increment()
+
+
+def enter_user_request_generator():
+    if _GRPC_ENABLE_FORK_SUPPORT:
+        _fork_state.active_thread_count.decrement()
+
+
+def return_from_user_request_generator():
+    if _GRPC_ENABLE_FORK_SUPPORT:
+        _fork_state.active_thread_count.increment()
+        block_if_fork_in_progress()
+
+
+def get_fork_epoch():
+    return _fork_state.fork_epoch
+
+
+def is_fork_support_enabled():
+    return _GRPC_ENABLE_FORK_SUPPORT
+
+    
+def fork_register_channel(channel):
+    if _GRPC_ENABLE_FORK_SUPPORT:
+        _fork_state.channels.add(channel)
+
+
+def fork_unregister_channel(channel):
+    if _GRPC_ENABLE_FORK_SUPPORT:
+        _fork_state.channels.remove(channel)
+
+
+class _ActiveThreadCount(object):
+    def __init__(self):
+        self._num_active_threads = 0
+        self._condition = threading.Condition()
+
+    def increment(self):
+        with self._condition:
+            self._num_active_threads += 1
+
+    def decrement(self):
+        with self._condition:
+            self._num_active_threads -= 1
+            if self._num_active_threads == 0:
+                self._condition.notify_all()
+
+    def await_zero_threads(self, timeout_secs):
+        end_time = time.time() + timeout_secs
+        wait_time = timeout_secs
+        with self._condition:
+            while True:
+                if self._num_active_threads > 0:
+                    self._condition.wait(wait_time)
+                if self._num_active_threads == 0:
+                    return True
+                # Thread count may have increased before this re-obtains the
+                # lock after a notify(). Wait again until timeout_secs has
+                # elapsed.
+                wait_time = end_time - time.time()
+                if wait_time <= 0:
+                    return False
+
+
+class _ForkState(object):
+    def __init__(self):
+        self.fork_in_progress_condition = threading.Condition()
+        self.fork_in_progress = False
+        self.postfork_states_to_reset = []
+        self.fork_handler_registered_lock = threading.Lock()
+        self.fork_handler_registered = False
+        self.active_thread_count = _ActiveThreadCount()
+        self.fork_epoch = 0
+        self.channels = set()
+
+
+_fork_state = _ForkState()