Execution of _pytest.hookspec.pytest_runtest_teardown is delayed

[pytestbot]

Originally reported by: Dave Hunt (BitBucket: davehunt, GitHub: davehunt)

---

It appears that since <> our plugin's pytest_runtest_teardown is often delayed in execution to the point of causing test failures.

We run Selenium tests using Selenium Grid for parallel execution. Since the above mentioned changeset if a test has to wait for an available Selenium node then the pytest_runtest_teardown is not executing until after the test session is timed out by Selenium. This means that the teardown method will fail as it attempts to use the session that has just timed out.

I can replicate this 100% of the time with this changeset. Please let me know how I can assist with resolving this issue.

Thanks in advance.

---

• Bitbucket: https://bitbucket.org/pytest-dev/pytest/issue/93

[pytestbot]

Original comment by holger krekel (BitBucket: hpk42, GitHub: hpk42):

---

I have a rough idea of what the problem might be but i am not 100% sure yet. Can you either provide a minimal example or otherwise describe in more detail what you are doing in runtest_setup and runtest_teardown and what in the actual test function? Is the test function the one that times out? (You don't happen to have some docs of your plugin?) Thanks, holger

[pytestbot]

Original comment by Dave Hunt (BitBucket: davehunt, GitHub: davehunt):

---

I can certainly work on creating a minimal example, however in the meantime I'll attempt to describe what we're doing...

The plugin [1] requests a Selenium session in pytest_runtest_setup. This new session request goes to Selenium Grid, which starts the session on a registered Selenium node. If there are no available Selenium nodes then the session request is queued in Grid until one is available.

The main test method basically sends various Selenium commands sent to the current session in order to automate the browser and check the AUT.

After the test method, the plugin captures some debug from the session in pytest_runtest_teardown and ultimately stops the session.

What appears to be happening is the main test method is completing, but the pytest_runtest_teardown is not executing when expected. After a timeout (default 3 minutes) the Selenium session is automatically terminated, and this appears to trigger the teardown. A potentially important observation is that the issue only occurs when Selenium Grid has to queue a session request.

There's no documentation for the plugin, but it is relatively simple and open source on github [1]. There is a good chance that I've misunderstood some things about pytest and pytest-xdist when developing this plugin, so feel free to pull it apart! :)

[1] https://github.com/davehunt/pytest-mozwebqa

[pytestbot]

Original comment by Dave Hunt (BitBucket: davehunt, GitHub: davehunt):

---

Simplified example here: https://gist.github.com/1416716

[pytestbot]

Original comment by holger krekel (BitBucket: hpk42, GitHub: hpk42):

---

thanks. I have an experimental fix. Can you try "pip install -i http://pypi.testrun.org -U pytest-xdist"? Should give 2.2.1.dev1 and 1.8.dev1 respectively

[pytestbot]

Original comment by Dave Hunt (BitBucket: davehunt, GitHub: davehunt):

---

Thanks Holger, that fix works for me.

[pytestbot]

Original comment by holger krekel (BitBucket: hpk42, GitHub: hpk42):

---

fix issue93 - avoid "delayed" teardowns for distributed testing by
simplifying handling of teardowns.