I recently upgrade my hadoop 0.2.203 to hadoop 1.1.1. In 0.2.203, I used the tar file and run all component under a single user which had password-less ssh across all nodes with no problems, but I used the debian package to install the 1.1.1 version. the installer created two user hdfs,mapred and their groups plus an extra group named hadoop. getting my clues from the installer I ran the namenode and datanodes under hdfs and jobtracker and tasktrackers under mapred. but now when i try to submit job under another user (the user that exist under all nodes) i get permission errors (for file created during staging). when i try to run the jobtracker under another user beside mapred i get this error:
Starting Apache Hadoop Job Tracker server: hadoop-jobtrackerstart-stop-daemon: unable to set gid to 202 (Operation not permitted)
so I was wondering, what is the correct or recommended configuration about under which and how many users is it best to run the services (namenode,datanode,jobtracker,tasktracer,...) and what set of permission and group membership should each of those user have, and what user would be allowed to submit jobs?