Refer to JIRA 1928 - https://issues.apache.org/jira/browse/MAPREDUCE-1928
The JIRA proposed the option of adding input on the fly to a job that has been submitted and may as well be in a running stage. The JIRA was implemented on Hadoop-20.2 version.
With the support for such a feature in Hadoop ( after application of the patch ), Hive can use the feature to optimize LIMIT queries that do not have an ORDER BY. For each query that qualifies to be of this kind, Hive needs to set appropriate parameters in the corresponding JobConf instance that gets created. The JobConf must have the attribute "dynamic.job" set to true and should have an appropriate InputProvider set. The input provider for optimizing the LIMIT query has been provided as part of the patch on Hadoop.