Queueing tasks replaces rejections, but the system still pushes work to workers

The author argues that a full worker pool only reports that no capacity is available because nothing records the work still waiting. Placing tasks into a queue and reporting each user's position turns rejections into waiting, which is a better experience at peak load. However, the scheduler still checks each worker for availability and then assigns tasks, which keeps the race condition and the cost of polling. The author notes that true pull-based scheduling, where free workers take tasks from the queue themselves, is needed to remove those problems. The post also points out that a VIP flag is passed along but never used by the scheduler.