Skip to main content

Asyncio wait_for that can handle simultaneous cancellation and future completion.

Project description

wait_for2

Alternate implementation of asyncio.wait_for(). It handles several edge cases like simultaneous cancellation of wait and completion of future differently and consistently across Python versions 3.7+.

Updates

This library may become unnecessary for Python 3.12+, see https://github.com/python/cpython/pull/28149#issuecomment-1560278644

A few more behaviour variances have been introduced that are not reflected in the details below. For example Python 3.9.10 changes the behaviour of simultaneous timeout and completion compared to previous 3.9 releases. PyPy 3 used to mirror the 3.7 behaviour, but the current release have changed the behaviour at some unspecified release to 3.9.10+.

Details

The tests in the repository are set up with TOX to cover and assert the following behaviours of wait_for and the alternate implementation for each Python version.

Cancellation behaviour with simultaneous result

Builtin asyncio.wait_for() behaviours:

  • Python 3.7: Cancellation of wait_for could lose the completed future's result.
  • Python 3.8+ and PyPy3: Cancellation of wait_for could lose the cancellation request.

Whenever waiting for a future's result the user expects to either have the future completed or cancelled. Unfortunately due to technical details there is a chance that both will happen simultaneously. The builtin version of asyncio.wait_for() shipped with Python either handles one or the other only. If losing the future's result or ignoring the cancellation is critical to the application it may not be suitable for use.

Using this implementation, in case both conditions occur at the same time a subclassed CancelledError will be raised which also contains the result of the future. The caller code must catch this exception and handle the result if it is important. Otherwise, it can be used the same way as the builtin wait_for.

If the caller prefers to handle the race-condition with a callback, the race_handler argument may be provided. It will be called with the result of the future when the waiter task is being cancelled. Even if this is provided, the special error will be raised in the place of a normal CancelledError.

NOTE: CancelledWithResultError is limited to the coroutine wait_for is invoked from prior to Python 3.11! If the wait_for is wrapped in tasks those will not propagate the special exception, but raise their own CancelledError instances. The callback-based solution may be preferred as that will always work in all Python versions.

This table summarizes the behaviours in the race-condition cases. The cross cells show what behaviour is observed:

  • LR: looses result (returned result or raised exception)
  • LC: looses cancellation request
  • RH: race-condition handling supported, cancellation is never ignored by the wait-for
Python 3.7 Python 3.8+ and PyPy3 wait_for2
explicit cancel & result (or exc.) LR LC RH

Timeout handling behaviour with results

When the timeout is reached, the inner future is cancelled. This can also cause race condition where the result is lost.

The cross cells show what behaviour is observed:

  • TE: prioritizes raising TimeoutError, looses result or exception
  • PR: prioritizes returning or raising the result exception
Python 3.7, 3.8 Python 3.9 Python 3.10+ and PyPy3 wait_for2
result after cancel by timeout TE TE PR PR
exception after cancel by timeout TE PR PR PR

Cancellation behaviour with timeout handling

Additionally, this implementation will inherit the behaviour of the inner future when it comes to ignoring cancellation. The builtin version prefers to always be cancellable, even if that means the wrapped future may not be terminated with it. (behaviour of builtin _cancel_and_wait) This behaviour is also improved in timeout-cancel edge cases, where the builtin would not wait for the termination of the inner future if the waiter was cancelled after timeout handling had already started. This is more consistent as the inner future must always be stopped for it to return.

The leftmost column describes the case where the behaviour is tested. It shows what timeout the wait-for is called with and when the cancellation occurs relative to it.

The cross-cells show the raised result and if the inner future is terminated before the wait-for implementation returns:

  • C/T: CancelledError/TimeoutError was raised
  • B/U: bound means the inner future terminated before the wait-for returned/unbound means the inner future was still running when the wait-for returned

The cells where the desired behaviour is observed (IMO) are formatted to be bold-italic.

Python 3.7 Python 3.8+ and PyPy3 wait_for2
no timeout, cancel C B C B C B
zero timeout, cancel before T U C U C B
zero timeout, cancel after T U C U C B
zero timeout, no cancel T U T B T B
some timeout, cancel before C U C B C B
some timeout, cancel after C U C U C B
some timeout, no cancel T B T B T B

Install & usage

A package is available on PyPI:

$ python -m pip install wait_for2
import asyncio
import wait_for2

task = asyncio.create_task(...)

def process_result(r, is_exc=False):
    print("processed:", r, is_exc)

...

try:
    process_result(await wait_for2.wait_for(task, 5.0))
except wait_for2.CancelledWithResultError as e:
    # NOTE: e.result could be an exception raised by the task; handling or ignoring it is up to the user code here
    process_result(e.result, e.is_exception)
    raise asyncio.CancelledError()

# alternatively with a callback:
process_result(await wait_for2.wait_for(task, 5.0, race_handler=process_result))

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

wait_for2-0.3.2.tar.gz (16.7 kB view details)

Uploaded Source

File details

Details for the file wait_for2-0.3.2.tar.gz.

File metadata

  • Download URL: wait_for2-0.3.2.tar.gz
  • Upload date:
  • Size: 16.7 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/3.7.1 importlib_metadata/4.9.0 pkginfo/1.8.2 requests/2.26.0 requests-toolbelt/0.9.1 tqdm/4.62.3 CPython/3.9.12

File hashes

Hashes for wait_for2-0.3.2.tar.gz
Algorithm Hash digest
SHA256 93863026dc35f3471104ecf7de1f4a0b31f4c8b12a2241c0d6ee26dcc0c2092a
MD5 551a1a2c19b676c43b6983195a6c7471
BLAKE2b-256 9126430de0d8aaf5aad9f23e83484e35a37a1426a33c2dd356b0d2a45d2fb552

See more details on using hashes here.

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page