If you misconfigure an aggregator such that it can't be loaded, it will cause a tablet to indefinitely attempt to minor/major compact. If you attempt to delete the table with this problem the tablet will never break out of the compaction attempt to a point where it will detect it needs to delete. The only way to break out of the loop is kill any tservers who are hosting tablets attempts to be compacted. This definately exists in 1.3, fairly confident it affects the other versions as well. We would need to have the compaction loops check somehow to detect if the tablet needs to be deleted before reattempting.
Additionally, a functional test should be written to exercise this error to prevent regression.