ViewFileSystem default ugi on mount point

[lohit <lo...@gmail.com>]

Hi Devs,

ViewFileSystem by default tries to populate ugi information of mount point
based on the user who is running the command. For example if I execute
'hadoop fs -ls /' on '/' which happens to be a ViewFileSystem mountpoint,
then all mountpoint under '/' would who the permission and owner as 'lohit'

For example this check is coming from ViewFileSystem.java.

public FileStatus getFileStatus(Path f) throws IOException {
      checkPathIsSlash(f);
      return new FileStatus(0, true, 0, 0, creationTime, creationTime,
          PERMISSION_RRR, ugi.getUserName(), ugi.getGroupNames()[0],

          new Path(theInternalDir.fullPath).makeQualified(
              myUri, null));



In our clusters we do not create user accounts on worker nodes. Now if any
Map or Reduce code does getFileStatus on mountpoint they fail when it tries
to fetch ugi.getGroupNames()[0] with NPE. One option we are thinking is to
populate this with 'nobody':'nogroup' kind of string if we cannot resolve
to user/group name instead of failing with NPE. What do you guys think
about this approach. Any other ideas/suggestion?

-- 
Have a Nice Day!
Lohit

Re: ViewFileSystem default ugi on mount point

[Juan Carlos <jc...@redoop.org>]

What about setting a ldap to have centralized user accounts?


2014/1/13 lohit <lo...@gmail.com>

> Hi Devs,
>
> ViewFileSystem by default tries to populate ugi information of mount point
> based on the user who is running the command. For example if I execute
> 'hadoop fs -ls /' on '/' which happens to be a ViewFileSystem mountpoint,
> then all mountpoint under '/' would who the permission and owner as 'lohit'
>
> For example this check is coming from ViewFileSystem.java.
>
> public FileStatus getFileStatus(Path f) throws IOException {
>       checkPathIsSlash(f);
>       return new FileStatus(0, true, 0, 0, creationTime, creationTime,
>           PERMISSION_RRR, ugi.getUserName(), ugi.getGroupNames()[0],
>
>           new Path(theInternalDir.fullPath).makeQualified(
>               myUri, null));
>
>
>
> In our clusters we do not create user accounts on worker nodes. Now if any
> Map or Reduce code does getFileStatus on mountpoint they fail when it tries
> to fetch ugi.getGroupNames()[0] with NPE. One option we are thinking is to
> populate this with 'nobody':'nogroup' kind of string if we cannot resolve
> to user/group name instead of failing with NPE. What do you guys think
> about this approach. Any other ideas/suggestion?
>
> --
> Have a Nice Day!
> Lohit
>