potiuk edited a comment on pull request #19860:
URL: https://github.com/apache/airflow/pull/19860#issuecomment-982552894
OK. @ashb @ephraimbuddy @uranusjr - I think I have a final confirmation that
this is the deferred reload from "spawn" test.
When I moved the `test_scheduler_multiprocessing_with_spawn_method` after
`test_scheduler_keeps_scheduling_pool_full` the "pool_full' tests stopped
failing, but the failures moved few tests down:
* test_more_runs_are_not_created_when_max_active_runs_is_reached:
*
https://github.com/apache/airflow/runs/4365506164?check_suite_focus=true#step:6:3949
* test_retry_still_in_executor:
*
https://github.com/apache/airflow/runs/4365506225?check_suite_focus=true#step:6:3987
*
https://github.com/apache/airflow/runs/4365505499?check_suite_focus=true#step:6:4408
* test_verify_integrity_if_dag_failed:
*
https://github.com/apache/airflow/runs/4365505550?check_suite_focus=true#step:6:4384
In 3 of the 4 tests (the first is exception) you can see the same side
effect (even more likely than before coming from the `spawned` test):
```
{settings.py:52} INFO - Configured default timezone Timezone('UTC')
```
I think in the first case the "log" entry was simply printed during the
tearDown of the previous tests.
And I have a theory that could explain why reloading MIGHT cause the problem
(or big part of it). @uranusjr - this is similar to the equality test problem
you solved recently in #19878 .
My theory is that while reloading settings, the Model classes get reloaded
and they are different `class` objects than they were a while ago. SQLAlchemy
when runnig SerializedDagModel.get() will find the objects in session only if
they are the same Class. But when the reload happens, the `dag_maker` for
whatever reason will use the new `SerializedDagModel` class - which is
different than `SerializedDagModel` when the model is queried with (this class
object is likely already loaded before and not reloaded).
Now - I still have no idea why spawned dag processor reload might cause it
as it shoudl be a different process and different python interpreter, I can
disable the `spawn` test case or separate it to be run completely
independently, but it would be great to find out if similar side effects can
happen outside of test environment.
--
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]
