Hi, I am running the 2.6.16.20 kernel what is otherwise a Debian Sarge system. I have two identical SATA hard drives in the system. Both have an identical boot partion at the start of the disks (/dev/sda1, /dev/sda2) and the remainder of the disks is used as RAID-1 on which I have LVM for my root partition and some other partitions. dromedary:~# uname -a Linux dromedary 2.6.16.20.rwl2 #1 Wed Jul 26 12:52:43 BST 2006 i686 GNU/Linux dromedary:~# lvm version LVM version: 2.02.14 (2006-11-10) Library version: 1.02.12 (2006-10-13) Driver version: 4.5.0 dromedary:~# mdadm --version mdadm - v2.5.5 - 23 October 2006 dromedary:~# As part of my backup process, I snapshot the data LV and make a copy of the snapshot to another machine's disk. I take an SHA1 checksum of the snapshot so that I can verify its accuracy. During this process, the results of the SHA1 checksum of the shapshot was changing between two values. I tracked this down to a single bit in the 10GB image which was returning 1/0, apparently randomly. Checking the "/proc/mdstat" suggested that the RAID array was intact as did a more detailed check using "mdadm --detail /dev/md0", so I spent a long time digging through the LVM configuration to try to find the problem. After much digging, I found that by running the RAID-1 array in degraded mode with one disk, the erratic behaviour stopped and I got a consistent value returned. When I rebooted the system with just the other disk, so again running the RAID-1 array in degraded mode, the erratic behaviour stopped but I got the other value returned. This indicated that the problem was that the RAID array had inconsistent data on its disks and was returning either disk's value when a read request was made, probably whichever got returned first... I have since rebuilt the RAID array by marking one of the disk as faulty then adding it as a spare again, thus causing the array to rebuild. The erratic behaviour has now stopped and everything is working properly again. This confused me, as I was (prehaps wrongly) expecting that a RAID-1 array would detect not just hard disk errors but also soft disk errors where the array returned inconsistent data from the two disks. After more digging, I have found that it is considered "good practice" to regularly ask the array to check itself using the "echo check > /sys/block/md0/md/sync_action" and then check the "/proc/mdstat" and "/sys/block/mdX/md/mismatch_cnt" files to determine the results. Is it in any way possible to make the array check both disks in the RAID-1 array when any data is read so that the returned data is verified "live" and then automatically raise a warning somehow? I appreciate that this would cause a reduction in performance, but I am willing to have that in exchange for increased robustness and immediate notification of a disk problem. Thanks, Roger - To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
