Thanks for your reply! 2011/6/20 Jan Kara <jack@xxxxxxx>: > Hum, if I understand your problem right, you should just mount the > filesystem with errors=remount-ro and you will get the behavior you need. > Or what is insufficient on that solution? Your patch surely provides more > flexibility but is that really needed? > 1.There are more than ten hard disks in each of our production machine, so it is not right for making the whole system panic, only based on one error in one harddisk. 2.There may be multiple tasks which access the same hard drive at the same time, so it is not ideal for changing the system to readonly, only based on one error in one task, while other task may be killed. That's why we have a relaxed restrictions, only when the error counter is grower than our threshold, we change fs to readonly or panic. When a system has a dozen hard drives, each hard drive is running several tasks on time, this feature is a real demand. > BTW, in cluster environment (which Hadoop seems to be AFAIU) it is standard > to mount filesystem even with stricter errors=panic so that node is taken > off the grid as soon as some problem happens. Usually handling service > failover is simpler than handling uncertain state after a filesystem error. > -- Wang Shaoyan -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
