To submit a larger job, one or several map or reduce task often insufficient physical memory, the last job failed, but the job is large, is time-consuming and error, to adjust the memory Settings to run, is time-consuming, in addition, after the big memory Settings, the number of tasks in parallel will decrease, affect the calculation of overall efficiency, whether in a certain task because out of memory error fails, through code or what is the mode of dynamically adjusting the task memory Settings, perform again?