I thought there was some mention of HDFS. Sorry, misunderstood!
Rohit-------- Original message -------- From: Steve Varnau <[email protected]> Date: 08/14/2016 12:16 AM (GMT-05:00) To: Rohit Jain <[email protected]> Cc: dev <[email protected]> Subject: RE: Trafodion User and Group id (installation) Those would be database users, not unix users, right? So different mechanism. -Steve (from mobile) On Aug 13, 2016 10:04 PM, "Rohit" <[email protected]> wrote: So how would Grant / Revoke authorization on objects that span across nodes be enforced by Hive or Trafodion, for example? Rohit -------- Original message -------- From: Steve Varnau <[email protected]> Date: 08/13/2016 11:09 PM (GMT-05:00) To: [email protected] Subject: RE: Trafodion User and Group id (installation) Amanda, I thought someone with a bit more HDFS expertise might answer this, but I'll take a shot. As long as the file storage/permission checks happen locally, I don't think there is any problem with the users' IDs being different across a cluster. This is not true for something like NFS where multiple systems are sharing same filesystem. So question is how HDFS handles it. According to [1], the owner/group are stored as strings, not numeric IDs. And permission checks happen on the namenode. So I think it is not necessary for the numeric user IDs to be in synch across a cluster. [1] https://hadoop.apache.org/docs/current/hadoop-project-dist/hadoop-hdfs/HdfsPermissionsGuide.html --Steve > -----Original Message----- > From: Amanda Moran [mailto:[email protected]] > Sent: Friday, August 12, 2016 12:08 PM > To: dev <[email protected]> > Subject: Trafodion User and Group id (installation) > > Hi there All- > > Does the group id and user id need to be the same on all nodes? > > I tested with them being different on each node, and I don't see any > issues. > Example: > node-2: trafodion:x:509:488::/home/trafodion:/bin/bash > node-1: trafodion:x:503:503::/home/trafodion:/bin/bash > > > Seems to work great. Is there something I don't know about? > > Thanks! > > -- > Thanks, > > Amanda Moran
