A SELECT or COUNT query in Hive will be executed as a MapReduce job even if the queries are made against a small table or dataset. Imagine that you want to execute one of these queries which should only take a few seconds... for example, the situation where the set up and tear down of the Hadoop job probably takes longer than the actual work portion of the job. Also imagine that another user has a complex and long-running job already executing on the cluster. Bad news for your job. If you're using mostly Hadoop default settings for the YARN scheduling algorithm, it's possible that your simple job won't be executed until the other is finished.