Repository navigation
BaseSelectorEventLoop.sock_{recv,sendall}() don't remove their callbacks when canceled #74250
Description
Activity
abacabadabacaba commented
on Apr 13, 2017 abacabadabacabamannequinMannequinAuthorCode:
import asyncio as a import socket as s @a.coroutine def coro(): s1, s2 = s.socketpair() s1.setblocking(False) s2.setblocking(False) try: yield from a.wait_for(loop.sock_recv(s2, 1), 1) except a.TimeoutError: pass yield from loop.sock_sendall(s1, b'\x00') yield s1.close() s2.close() loop = a.get_event_loop() loop.run_until_complete(coro())
Result:
Exception in callback BaseSelectorEventLoop._sock_recv(<Future cancelled>, True, <socket.socke...2049, proto=0>, 1) handle: <Handle BaseSelectorEventLoop._sock_recv(<Future cancelled>, True, <socket.socke...2049, proto=0>, 1)> Traceback (most recent call last): File "/usr/lib/python3.6/asyncio/events.py", line 127, in _run self._callback(*self._args) File "/usr/lib/python3.6/asyncio/selector_events.py", line 378, in _sock_recv self.remove_reader(fd) File "/usr/lib/python3.6/asyncio/selector_events.py", line 342, in remove_reader return self._remove_reader(fd) File "/usr/lib/python3.6/asyncio/selector_events.py", line 279, in _remove_reader key = self._selector.get_key(fd) File "/usr/lib/python3.6/selectors.py", line 189, in get_key return mapping[fileobj] File "/usr/lib/python3.6/selectors.py", line 70, in __getitem__ fd = self._selector._fileobj_lookup(fileobj) File "/usr/lib/python3.6/selectors.py", line 224, in _fileobj_lookup return _fileobj_to_fd(fileobj) File "/usr/lib/python3.6/selectors.py", line 41, in _fileobj_to_fd raise ValueError("Invalid file descriptor: {}".format(fd)) ValueError: Invalid file descriptor: -1
See also https://bugs.python.org/issue34795
I think this broke asyncio a bit. These two uvloop sock_* API tests no longer pass on asyncio 3.8:
-
https://github.com/MagicStack/uvloop/blob/master/tests/test_sockets.py#L192
-
https://github.com/MagicStack/uvloop/blob/master/tests/test_sockets.py#L238
Andrew, since this was your patch, please take a look.
-
I'll elevate the status so that we don't forget before 3.8.1 is too close
I'll elevate the status so that we don't forget before 3.8.1 is too close
Andrew, Yury: ping. 3.8.1 cutoff is approaching.
Not sure if failed uvloop tests are correct.
The scenario is like the following:- Suppose we have an unblocking socket connected to peer.
- Create a task for reading data:
task = asyncio.create_task(loop.sock_read(sock, 1))
Note, the task is not started yet. - Just after it, cancel the task without waiting for the actual cancellation
task.cancel() - Finally, read from the socket again:
data = await loop.sock_read(sock, 1)
If I put context switch (await asyncio.sleep(0)) between 3) and 4) *or* replace direct sock_read() call in 4) with creation a task (data = await asyncio.create_task(loop.sock_read(sock, 1))) the cancellation of former read is performed and test passes.
I very doubt if any sane code is organizing like this test: start delayed reading, cancel it and read again.
The worse, neither previous not current sock_read() implementation doesn't prevent the concurrent reading which basically delivers data in an unpredictable order. Honestly, I'm not sure if we need to provide the concurrency guarantee for such low-level functions. The code built on top of these low-level primitives should handle the cuncurrent access problem IMHO.
Note: this is going to miss Python 3.8.1 as I'm releasing 3.8.1rc1 right now. Please address this before 3.8.2 (due in February).
27 remaining items
No problem at all, and thanks for clarifying.
If enlarging the timeout works in some of your cases, that really inspires me on a potential cause of the issue in the test. To simulate the race condition specifically for loop.sock_sendall(), I needed a connected socket with a full send buffer so that the next sock_sendall() call would block. To achieve this, I wrote something like this:
It might be that it cost too much time in these loops (especially the 2nd one because the logs showed that it was blocked on calling sock_sendall()). But that's just a guess yet, let me add some more debug logs and test locally.
That seems reasonable to me and would make the most sense at a glance. If this is confirmed to be the source of the issue, we should be able to just reduce the size of the socket's send buffer via SO_SNDBUF before sending to it so that it fills up much more quickly.
fantix commented
on Jun 14, 2020 fantixmannequinMannequinOK I think I solved the problem by improving the two loops using smarter buffer sizes:
# fill the buffer until sending 5 chars would block size = 8192 while size > 5: with self.assertRaises(BlockingIOError): while True: sock.send(b' ' * size) size = int(size / 2)
and
# receive everything that is not a space async def recv_all(): rv = b'' while True: buf = await self.loop.sock_recv(server, 8192) if not buf: return rv rv += buf.strip()
This test is now running ~25x faster, and the load test crashed my laptop a few times before it hits a timeout.
Last question before PR pls: given the fact that this test is to cover a fixed case when loop.sock_*() was hanging forever, should I keep the wait_for(..., timeout=10)?
fantix commented
on Jun 14, 2020 fantixmannequinMannequinand yes, updated the SO_SNDBUF too
Last question before PR pls: given the fact that this test is to cover a fixed case when loop.sock_*() was hanging forever, should I keep the wait_for(..., timeout=10)?
For now, I think we can keep the asyncio.wait_for() w/ 10 the sec timeout. I'll have to run the tests locally though w/ the patch to double check (once the PR is opened), and it may still end up needing to be increased if the test fails again in the future. It should be okay though.
Metadata
Metadata
Assignees
Projects
- Status
Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.
Show more details
GitHub fields:
bugs.python.org fields: