I would rather carry this over trunk as well.
We don't want to end up with different infrastructure depending on the branch. It may also be useful for other branches such as 0.22
NOTE: this is specifically designed for the work in hadoop-0.23 branch. We might or might not decide to carry this over to trunk.