At present, for a MapReduce Job Submission to work, the user ID submitting the job should be the same user ID that started the JobTracker and TaskTracker. We need to be able to support Job Submission by multiple users as the current user (mapred) will not always have permissions to access all the data for all foreseeable Hadoop jobs.
Multitenancy has now been addressed, with an upcoming ACL patch to be released
shortly.
The solution is to add a hook into the file system which, and based on a regex,
add ACL privileges for a hadoop / mapred daemon, so that it can always see any
files which are jobtracker related. This way, any user can submit a job and be
gauranteed that the jobtracker daemon (specified in a config parameter), will
be able to run that job.
See https://forge.gluster.org/hadoop/pages/Security