Linux raid5 trouble (please help, will pay)

dualblade

Supreme [H]ardness
Joined
Nov 19, 2000
Messages
4,180
ok, so i've done something really stupid - i've broken the golden rule that 'raid is not a backup'. i have an 8 drive raid5 volume on an ubuntu server running evms (the raid5 is a legacy volume, not an evms volume, but i use evms for management). the os drive (not one of the 8 drives) failed recently. i rebuilt the server with a new main drive, but when it came up, i was getting buffer errors on one of the raid drives. no big deal, as i can run a drive down on the array. when i tried to go into evms, 7 of the 8 drives have the following error:

GptSegMgr: Alternate GPT Header is missing or corrupt. Marking sdh_metadata2 dirty to correct the problem.

since only 7 of the 8 drives are plugged in (the one with buffer errors i unplugged), this means all the drives had their gpt headers scrambled. evms also has the error that the raid5 volume is corrupt (because of the gpt errors, the array is missing all its disks). i'm guessing the problem occured when the server crashed, but is there any way to fix these drives? also, i'm not positive that the disks were running gpt. it's been a while since i set up the array and i don't remember if i used gpt or mbr on the disks. i'm guessing that it was gpt, since the partition is >2tb. all drives had gpt errors, and i think it's unlikely that all would be misrepresented as gpt if they were mbr, but i figured i'd toss it out there

i've got some pictures that are really important to me on this array, so if anyone could help me out, it would be much appreciated. if someone wants to take this as a job, i'm willing to pay
 
Hmm... yeah EVMS was really cool until it sorta just died.

If this isn't EVMS volume and is only a legacy volume, or didn't they get called compatibility volumes or something? Anyways you shouldn't need EVMS, right? If you just boot off a LiveCD and try to use mdadm to assemble the array what does it say?

I mean step1 would be getting the array to assemble.... worst case is mdadm gives some error and you have to use force option and hope for best.

2nd step would dealing with corrupted partition table... this almost seems unlikely to have happened and maybe just EVMS being crazy. But if you do get array assembled and partition table is screwed you can install testdisk and it can search the array for partition info. Even if you wipe out a partition table testdisk should be able to scan the drive/array and figure out where the partitions start/end.

After that gets fixed just try to mount file system read only. Really how much data could have gotten corrupted?

I wouldn't take my advise.... hopefully someone else has some better ideas ;).
 
yes, compatibility volume is what i meant. i don't need evms, i just liked the ncurses interface over the straight command line of mdadm. the idea of a compatibility volume is that in case evms ever went away, i should be able to use other tools to make the array work.
 
if i use mdadm to assemble an array, do i tell it number of devices=8 but only list 7 devices (since one of the drives is getting buffer errors)?
 
You don't need that, assembling array only needs an array name and list of drives. Any additional information will be read off the drives.

Eg
mdadm -A /dev/md0 /dev/sda /dev/sdb /dev/sdc
 
thanks for the advice. i'll give it a shot and see how things go.

i'm not really sure what happened with evms. i'm also confused as to why no one else has come up with a replacement, as i can't imagine there's no demand for a convenient raid management interface
 
Back
Top