How should xdist be gracefully exited mid run when some exception occurs

[symonk]

I have a scenario where I am applying some logic during an xdist -n=X run on the master node and sometimes explicitly on GW0 (tho i might refactor that to a plugin and have master handle it). In the event that something has gone wrong, how should I gracefully exit pytest/xdist across all workers?

Is there a simple means of doing this? I'm achieving it through some hacky methods at the moment and I'm wondering if there is actually a proper mechanism for handling it

[nicoddemus]

I didn't try it, but maybe pytest.exit() will do the trick? Don't remember the details right now.

[symonk]

thanks, will have a try with it

[Raphasha27]

Use pytest-xdist with --max-worker-restart and a timeout plugin:

pytest -n auto --max-worker-restart=3 --timeout=60

For graceful cleanup on exceptions, wrap your test logic and call pytest.exit():

# conftest.py
import pytest
import sys

def pytest_sessionfinish(session, exitstatus):
    if exitstatus != 0:
        print("Gracefully shutting down xdist workers...")
        # workers drain naturally after sessionfinish

If a single worker crashes, xdist reassigns its tests to other workers. To handle long-running hangs, combine with pytest-timeout:

# pytest.ini
[pytest]
timeout = 120
timeout_method = signal  # or thread

This kills hung workers and lets remaining workers continue. For full control, use --dist loadscope to group tests by class/module so a single failure doesn't fragment test groups.