You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

168 lines
6.3 KiB
Python

import signal
from contextlib import contextmanager
from collections import OrderedDict
import trio
from ._util import signal_raise, is_main_thread, ConflictDetector
# Discussion of signal handling strategies:
#
# - On Windows signals barely exist. There are no options; signal handlers are
# the only available API.
#
# - On Linux signalfd is arguably the natural way. Semantics: signalfd acts as
# an *alternative* signal delivery mechanism. The way you use it is to mask
# out the relevant signals process-wide (so that they don't get delivered
# the normal way), and then when you read from signalfd that actually counts
# as delivering it (despite the mask). The problem with this is that we
# don't have any reliable way to mask out signals process-wide -- the only
# way to do that in Python is to call pthread_sigmask from the main thread
# *before starting any other threads*, and as a library we can't really
# impose that, and the failure mode is annoying (signals get delivered via
# signal handlers whether we want them to or not).
#
# - on macOS/*BSD, kqueue is the natural way. Semantics: kqueue acts as an
# *extra* signal delivery mechanism. Signals are delivered the normal
# way, *and* are delivered to kqueue. So you want to set them to SIG_IGN so
# that they don't end up pending forever (I guess?). I can't find any actual
# docs on how masking and EVFILT_SIGNAL interact. I did see someone note
# that if a signal is pending when the kqueue filter is added then you
# *don't* get notified of that, which makes sense. But still, we have to
# manipulate signal state (e.g. setting SIG_IGN) which as far as Python is
# concerned means we have to do this from the main thread.
#
# So in summary, there don't seem to be any compelling advantages to using the
# platform-native signal notification systems; they're kinda nice, but it's
# simpler to implement the naive signal-handler-based system once and be
# done. (The big advantage would be if there were a reliable way to monitor
# for SIGCHLD from outside the main thread and without interfering with other
# libraries that also want to monitor for SIGCHLD. But there isn't. I guess
# kqueue might give us that, but in kqueue we don't need it, because kqueue
# can directly monitor for child process state changes.)
@contextmanager
def _signal_handler(signals, handler):
original_handlers = {}
try:
for signum in set(signals):
original_handlers[signum] = signal.signal(signum, handler)
yield
finally:
for signum, original_handler in original_handlers.items():
signal.signal(signum, original_handler)
class SignalReceiver:
def __init__(self):
# {signal num: None}
self._pending = OrderedDict()
self._lot = trio.lowlevel.ParkingLot()
self._conflict_detector = ConflictDetector(
"only one task can iterate on a signal receiver at a time"
)
self._closed = False
def _add(self, signum):
if self._closed:
signal_raise(signum)
else:
self._pending[signum] = None
self._lot.unpark()
def _redeliver_remaining(self):
# First make sure that any signals still in the delivery pipeline will
# get redelivered
self._closed = True
# And then redeliver any that are sitting in pending. This is done
# using a weird recursive construct to make sure we process everything
# even if some of the handlers raise exceptions.
def deliver_next():
if self._pending:
signum, _ = self._pending.popitem(last=False)
try:
signal_raise(signum)
finally:
deliver_next()
deliver_next()
# Helper for tests, not public or otherwise used
def _pending_signal_count(self):
return len(self._pending)
def __aiter__(self):
return self
async def __anext__(self):
if self._closed:
raise RuntimeError("open_signal_receiver block already exited")
# In principle it would be possible to support multiple concurrent
# calls to __anext__, but doing it without race conditions is quite
# tricky, and there doesn't seem to be any point in trying.
with self._conflict_detector:
if not self._pending:
await self._lot.park()
else:
await trio.lowlevel.checkpoint()
signum, _ = self._pending.popitem(last=False)
return signum
@contextmanager
def open_signal_receiver(*signals):
"""A context manager for catching signals.
Entering this context manager starts listening for the given signals and
returns an async iterator; exiting the context manager stops listening.
The async iterator blocks until a signal arrives, and then yields it.
Note that if you leave the ``with`` block while the iterator has
unextracted signals still pending inside it, then they will be
re-delivered using Python's regular signal handling logic. This avoids a
race condition when signals arrives just before we exit the ``with``
block.
Args:
signals: the signals to listen for.
Raises:
TypeError: if no signals were provided.
RuntimeError: if you try to use this anywhere except Python's main
thread. (This is a Python limitation.)
Example:
A common convention for Unix daemons is that they should reload their
configuration when they receive a ``SIGHUP``. Here's a sketch of what
that might look like using :func:`open_signal_receiver`::
with trio.open_signal_receiver(signal.SIGHUP) as signal_aiter:
async for signum in signal_aiter:
assert signum == signal.SIGHUP
reload_configuration()
"""
if not signals:
raise TypeError("No signals were provided")
if not is_main_thread():
raise RuntimeError(
"Sorry, open_signal_receiver is only possible when running in "
"Python interpreter's main thread"
)
token = trio.lowlevel.current_trio_token()
queue = SignalReceiver()
def handler(signum, _):
token.run_sync_soon(queue._add, signum, idempotent=True)
try:
with _signal_handler(signals, handler):
yield queue
finally:
queue._redeliver_remaining()