I can observe the same problem with version 2.0.2:

• Tasks fail, because a DAG/task has gone missing (we are using dynamically created DAGs, and they can go missing)
• The scheduler keeps those queued
• The pool gradually fills up with these queued tasks
• The whole operation stops, because of this behaviour

My current remedy:

• Manually remove those queued tasks

My desired solution:

When a DAG/task goes missing while it is queued, it should end up in a failed state.

Originally posted by @lukas-at-harren in #13542 (comment)