This patch breaks hadoop for me. Can we reconsider it?
It is not safe to call
in order to close one filesystem connection.
It causes the parent JobTracker's FileSystem object to be closed as well, making it impossible (for me) to successfully submit a job.