There are a few problems with the current way we handle shutting down the pre-fetch functionality.
(1) Currently, it is possible for pre-fetch to accidentally leave entries on the pre-fetch on the pre-fetch queue as there is no logic in the __wt_prefetch_destroy function that drains the remaining entries on the queue.
(2) Pre-fetch should not be running during shutdown, but the current implementation does not stop the pre-fetch threads from running until very late into the __conn_close function.