June 1, 200917 yr Recovering the failed sectors on a second disk still requires the first failed disk. Yes, but as long as the relative unreadable sectors of the second failed disk are still readable on all other disks (including the first failed), and have not been written to, those failed sectors from the second disk could be recovered. People do have a choice if desired to write a script to turn off their array if a disk fails, then it will surely get noticed. ... In addition, I would prefer limetech spend time on implementing a RAID6 or hot spare option rather then switching to read only. True enough. I'll concede.
June 1, 200917 yr Recovering the failed sectors on a second disk still requires the first failed disk. Yes, but as long as the relative unreadable sectors of the second failed disk are still readable on all other disks (including the first failed), and have not been written to, those failed sectors from the second disk could be recovered. This is where I get confused... if there is a read failure on another disk for a certain amount of sectors, the drive will be taken off line causing the degraded array to fail out completely. OR The read will be reconstructed from each of the disks in the array including the virtual disk. (not sure... perhaps a temporary read error). readonly or not... If there is an error on another disk, you risk the degraded array failing. If you are in readonly mode, would you loose the ability for unRAID to update the failed sectors? People do have a choice if desired to write a script to turn off their array if a disk fails, then it will surely get noticed. ... In addition, I would prefer limetech spend time on implementing a RAID6 or hot spare option rather then switching to read only. True enough. I'll concede. It was worthy of a discussion to debate the merits. I think we are in agreement, the time spent for such a feature might be better spent on improved redundancy or enhanced notification. The possibility still exists for those wanting this feature to remount the filesystems readonly in some sort of monitoring script.
June 1, 200917 yr This is where I get confused... if there is a read failure on another disk for a certain amount of sectors, the drive will be taken off line causing the degraded array to fail out completely. OR The read will be reconstructed from each of the disks in the array including the virtual disk. (not sure... perhaps a temporary read error). Without modification, unraid may not be able to recover the data on its own. You would likely need to script the recovery of the failed sectors yourself -- but the data is there, and recoverable, as long as all disks but one can be read for a given block, and that block was not modified by a write after a failed disk was taken offline. I'll try a visual example: 0 = Good, readable block X = Bad, unreadable block block - 1 2 3 4 5 6 7 8 9 0 parity: 0 0 0 0 0 0 0 0 0 0 disk1 : 0 0 0 0 X X X 0 0 0 - Not yet detected as failed. disk2 : X X X X 0 0 X 0 X 0 - Unraid detected as failed and took offline. disk3 : 0 0 0 0 0 0 0 0 0 0 As long as blocks 5 and 6 were not written to on disk3, the data on block 5 and 6 from disk1 can be reconstructed using the data from the other disks. Block 7 on disk1 and disk2 would be permanently lost in this example. Writing to blocks 5 and 6 on disk3 or the simulated disk2 would result in permanent loss of the data on blocks 5 and 6 of disk 1. I think we are in agreement, the time spent for such a feature might be better spent on improved redundancy or enhanced notification. Agreed.
June 2, 200917 yr Writing to blocks 5 and 6 on disk3 or the simulated disk2 would result in permanent loss of the data on blocks 5 and 6 of disk 1. On second thought, the above assertion is wrong... only a write to blocks 5 and 6 of the simulated, offline, disk2 would cause blocks 5 and 6 on the also failing disk1 to become unrecoverable. Sorry for the exposure to my brain fart.
June 2, 200917 yr Writing to blocks 5 and 6 on disk3 or the simulated disk2 would result in permanent loss of the data on blocks 5 and 6 of disk 1. On second thought, the above assertion is wrong... only a write to blocks 5 and 6 of the simulated, offline, disk2 would cause blocks 5 and 6 on the also failing disk1 to become unrecoverable. Sorry for the exposure to my brain fart. If you're looking for me, I'm the one sitting in the corner wearing the funny cap. unraid would not be able to write to the simulated disk2 without first being able to read the damaged blocks of disk1. I completely concede my point of a read only degraded array allowing a chance of more data recovery. Unraid naturally wouldn't be able to write to, and thus corrupt, such blocks in a degraded array. Boy -- that was embarrassing.
June 2, 200917 yr If you're looking for me, I'm the one sitting in the corner wearing the funny cap. You had me going there for a while.. I could not quite wrap my brain around it. What I did think was an interesting idea was the possibility of a readonly flag that could be placed on the user share to force any future user share allocation to bypass this disk. This in effect would force any new writes or cache moves to slide to the next disk in the user share, thus saving the possibility of "virutal" disk loss if a second disk fails. I'm still in the camp that time spent on raid6 alleviates this need. (still an interesting debate).
June 2, 200917 yr Here's an idea. Catastrophic recovery mode. You have 2 disks that unRAID has disabled (suspect disks). Array is offline. If the drives have not DIED, but just had a write error, as noted before it is unlikely the exact same sector went bad on both drives, so parity can be used to do a 100% recovery. Approach 1: Install 2 new drives, equal or greater in size to each "suspect" disk. Then "pair" each new disk to one suspect disk. After clearing the disks, unRAID then copies the data from each suspect disk to it's pair. Recovery is attempted for any sector errors.... but no write-back to the original disk and no disabling of the disk. Then the suspect disks are removed from the array schema and replaced by the new disks. Approach 2: Install ONE new disk (outside the array) that is as big as the two suspect disks combined, and just copy all the data from both suspect drives to it, attempting sector recovery for any read errors. Then shutdown and physically remove the suspect disks. Of course, RAID6 would still be better.
August 31, 200916 yr I'm reading this thread and I know it's old, but I think I've come up with another way to recover: Create a Self Healing Feature as described here: http://lime-technology.com/forum/index.php?topic=4269.0 I'm looking for some feedback. Thanks, KermitJr
Archived
This topic is now archived and is closed to further replies.