potiuk commented on PR #30588:
URL: https://github.com/apache/airflow/pull/30588#issuecomment-1511512047
Going back to the PR that caused test failures on our self-hosted runners
after succeeding on public runners.
https://github.com/apache/airflow/pull/30362 (which we had to revert).
This is an interesting one.
I figured that one @vincbeck @shahar1 @uranusjr . I know why the tests were
failing on self-hosted, I am not 100% sure though why they were not failing in
the public runners, but I have a theory.
The problem was a side effect of the new `test_except_value_error` (from
`tests/operators/test_python.py`) on few other tests, namely monkeypatching the
loads method of pickle library with mock throwing DeserializationErro:
```` python
def test_except_value_error(self):
...
task.pickling_library.loads =
mock.Mock(side_effect=DeserializingResultError) # <- THIS was the problem
with pytest.raises(DeserializingResultError):
task._read_result(path=mock.Mock())
````
I changed it to this and it fixed the problem:
```python
@mock.patch("pickle.loads") # <- this will restore the monkey patched
method after test completes
def test_except_value_error(self, loads_mock):
...
loads_mock.side_effect = DeserializingResultError
with pytest.raises(DeserializingResultError):
task._read_result(path=mock.Mock())
```
The thing was that picklling library is a module - either `pickle` or `dill`
and monkeypatchign `loads` library there, monkeypatched it also for any of the
tests run within the same interpreter. Default is `pickle`, so subsequent
tests that used the same "pickle" library used the same monkeypatched loads
method that threw DeserializationError.
Now, I am not 100% sure why it has not failed on public runners. But I have
a theory.
The way how `pytest` manage the modules and whether it keeps them in memory
between the tests is a bit of mystery to me (I've hit a few times somewhat
strange behaviour with objects kept in memory), but a theory I have is that if
the tests were run in memory constrained environment, the loaded module with
monkeypatched method was removed before the "test_external_python" was run in
the "decorators" pckages run - that would explain why the tests were not
failing in Public runners.
Anyone has another theory ?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
