Make a separate work queue to track async operations. Keep separate work queues for links and nodes. This avoids lockups when some async operation take a long time and the work queue has SYNC_WAIT operations.
