Do we need for more flexibility in the way tasks are run ?

[Brice Arnould]

Hi !
I would like your advice about an proposing I was going to create on 
Jira when I realised that I might be completely wrong ^.^

So, the context is that, with [HADOOP-3421] speaking about sharing a 
cluster among more than one organization (so potentially with 
non-cooperative users), and posts here speaking about virtualization and 
the ability of re-using the TaskTracker's VM to run new tasks, I told 
myself that it could be useful to choose the way TaskRunners run their 
children.

More specifically, it could be useful to at least provide a way to 
imprison a Task in its working directory, or in a virtual machine.
The usefulness of sharing the TaskTracker VM seems more arguable, but it 
might still be interesting.

So I started writing a prototype, by extracting and generalizing parts 
of the TaskRuner, like I tried before with the JobTracker. But then I 
realized that, if we don't need more than imprisoning, it could be much 
easier to just provide a class inheriting of ShellCommandExecutor that 
would run its commands in a sandbox.

So I am asking for your opinions if my first idea was a good one, or if 
I should re-work the second one... or drop both :-P
Thanks !
Brice