On an EMR cluster, we have a situation where most users can run Hive queries (against tables based in S3), but some users cannot. Those users which cannot get a curious S3