Test for #26804 has a tendency to fail on Oracle databases

[Jensen Cochran]

According to comment on the pull request for #26804 (​https://github.com/django/django/pull/6831), the test for the issue (written by me) has a tendency to fail on Oracle databases. The test assumes that the database will take no longer than 50ms to lock the associated row, but it seems that this assumption is incorrect in some cases.

I have written a patch for this that should mostly fix the issue. Instead of assuming the database has locked the row, I set a value in the spawned thread after select_for_update has ran and poll that value in the main thread.

However, there still has to be some kind of timeout as we cannot wait forever for the database to lock the row. I have set that timeout at 0.5 seconds and have changed the thread/callable sleep time to 0.5 seconds as well to allow for more room for slow databases. There is no reason to have the timeout be any longer than the thread sleep time, because once the main thread waits that long the assertion would pass no matter what anyway. If the timeout expires the test is skipped.

Alternatively, I could fail if the timeout occurs, but this has potential to fail somewhat randomly as it is doing now.

If update_or_create were ever changed such that the locking behavior was removed, the test would fail as expected so it should still serve its purpose.