You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
wipe_task now happens in complete task, the moral successor of handle_app_update
but it now happens *before* the user observes a task complete.
this makes the task be able to be removed from the tasks table before the
user observes the future is complete.
you might say thats a new race condition: the user cannot see the task table entry
at the moment of completion. but that was never strongly guaranteed: the callback
to remove the entry could happen before or after the user observed the completion.
now theres a stronger assertion: it will definitely happen before the user observes
task completion via an AppFuture.
TODO: are there already tests about this?
When i removed wipe_task calls entirely, no per-config test failed...
parsl/tests/test_python_apps/test_garbage_collect.py
- this test was very slightly racey before and this was noticeable
in some race-condition fuzzing work I've done
0 commit comments