What will happen if the output directory already exists for a MapReduce job?