Fix flaky tests in TestChildProcessCleanup (test_stdio.py)

[maxisbey]

Initial Checks

• I confirm that I'm using the latest version of MCP Python SDK
• I confirm that I searched for my issue in https://github.com/modelcontextprotocol/python-sdk/issues before opening this issue

Description

Three tests in TestChildProcessCleanup are flaky. They fail intermittently on macOS (darwin) with the same root cause: the child process hasn't started writing to the marker file before the assertion checks.

Affected tests:

• TestChildProcessCleanup.test_basic_child_process_cleanup
• TestChildProcessCleanup.test_nested_process_tree
• TestChildProcessCleanup.test_early_parent_exit

CI evidence: Tests observed failing on 2025-11-12 and continue to be flaky.

Output:

________ TestChildProcessCleanup.test_basic_child_process_cleanup ________
[gw9] darwin -- Python 3.11.14

    async def test_basic_child_process_cleanup(self):
        ...
        # Verify child is writing
        if os.path.exists(marker_file):  # pragma: no branch
            initial_size = os.path.getsize(marker_file)
            await anyio.sleep(0.3)
            size_after_wait = os.path.getsize(marker_file)
>           assert size_after_wait > initial_size, "Child process should be writing"
E           AssertionError: Child process should be writing
E           assert 0 > 0

tests/client/test_stdio.py:310: AssertionError

________ TestChildProcessCleanup.test_nested_process_tree ________
[gw9] darwin -- Python 3.11.14

>           assert new_size > initial_size, f"{name} process should be writing"
E           AssertionError: parent process should be writing
E           assert 0 > 0

tests/client/test_stdio.py:413: AssertionError

________ TestChildProcessCleanup.test_early_parent_exit ________
[gw9] darwin -- Python 3.11.14

>           assert size2 > size1, "Child should be writing"
E           AssertionError: Child should be writing
E           assert 0 > 0

tests/client/test_stdio.py:494: AssertionError

Python & MCP Python SDK

Python 3.11.14, macOS (darwin)
Latest SDK version

_{AI Disclaimer}

[BabyChrist666]

Opened a PR to fix this: #2071

The fix replaces the fixed-duration anyio.sleep() calls with a _wait_for_file_growth() polling helper that waits until the marker file exists and is actively growing, with a 10s timeout. This eliminates the race condition where assertions fire before child processes have started writing on slow CI machines.