[Bug]: async_playwright() leaves blocking background task if cancelled quickly

[sattlerc]

Version

1.62.0

Steps to reproduce

With Pythin 3.14.7 and Playwright 1.62.0:

import asyncio

from playwright.async_api import async_playwright

async def run_browser():
    async with async_playwright():
        pass

async def run():
    task = asyncio.create_task(run_browser())
    await asyncio.sleep(0.01)  # for me: some value between 0.001 and 0.3
    task.cancel()
    try:
        await task
    except asyncio.CancelledError:
        pass

asyncio.run(run())

Expected behavior

The program exits.

Actual behavior

The program hangs.

Additional context

No response

Environment

Linux 7.1.3
Python 3.14.7
playwright 1.62.0 (latest version on PyPI)

[sreegithub89]

should/have to be async with async_playwright(): in try block?

[sattlerc]

should/have to be async with async_playwright(): in try block?

No. This raises a CancelledError as expected, which causes the task to be cancelled. The problem is that async_playwright().__aenter__() leaves behind some blocking background task.

[sreegithub89]

import asyncio
from playwright.async_api import async_playwright

async def run_browser():
    # Using async context manager ensures cleanup even if cancelled
    async with async_playwright() as p:
        browser = await p.chromium.launch()
        page = await browser.new_page()
        await page.goto("https://example.com")
        await browser.close()

async def run():
    task = asyncio.create_task(run_browser())
    await asyncio.sleep(0.01)  # simulate early cancellation
    task.cancel()
    try:
        await task
    except asyncio.CancelledError:
        # Context manager guarantees playwright.stop() is called
        print("Task cancelled safely, no background tasks left.")

asyncio.run(run())

[sattlerc]

That's essentially the same program. Hangs for me.

[sattlerc]

The async code doesn't seem easy to understand to me and possibly somewhat brittle.

One thing that immediately stands out: the Python documentation for create_task says:

Save a reference to the result of this function, to avoid a task disappearing mid-execution. The event loop only keeps weak references to tasks. A task that isn’t referenced elsewhere may get garbage collected at any time, even before it’s done.

There seem to be several places where that is not the case. For example, in PlaywrightContextManager.__aenter__:

        loop.create_task(self._connection.run())

[yury-s]

Cancelling a task that is running a Playwright call is not supported and results in undefined behavior. Documenting this in #42312. If you can reproduce a hang without task.cancel(), please open a new issue.

[sattlerc]

Cancelling a task that is running a Playwright call is not supported and results in undefined behavior. Documenting this in #42312. If you can reproduce a hang without task.cancel(), please open a new issue.

That would be unfortunate. It would mean that running playwright as a task inside a TaskGroup is undefined behaviour. This is how I found this issue: starting playwright in parallel with some other task that failed.