The only point #setAllowQueuedScheduling(false) is in JobGraphGenerator. IIRC we always #setAllowQueuedScheduling(true) after the generation and before the submission. For reduce confusion I propose to remove JobGraph#allowQueuedScheduling and refactor the related logic to all respect true.
This flag is originally used for configuring different resource allocation strategy between legacy mode and FLIP-6 arch. And there remains branches in Scheduler which might cause further confusion.