mirrored VG issue

[meekrob]

Hi All,

im having problem with one volume group vg01 (mirrored PVs config). The fact is that one of the disks went bad however, this logically does not explain that i cant have access to the vg01's file systems anymore as it mirrored.

vgdisplay -v vg01 showed that one PV is unavailable and the other available with a stale status for all corresponding lvols.

The weird thing is that i cant access the data on the related filesystems and i lost ssh / telnet access to the server even after restarting related net daemons... (no modifications were made to corresponding system files). In addition, and whenever i issued bdf it freezes and did not return the prompt unless i press Ctrl C.

Ive replaced the defective disk with a new one and did the vgcfgrestore / vgsync and tried to mount all file systems without success.  I also tried fsck as it was showing "invalid or bad super block message ..." and it suggests retrieving an auxiliary super block without success

I pulled out the 2 VG01 disks and when i restarted the server, ssh / telnet went back but i lost all VG01.

How can that be? and what is the mirroring used for? in addition, is there a way to get this data on the remaining disk for VG01?

Thanks in advance for your suggestions

[Torsten.]

>> How can that be?

Normally not possible.

But I have seen configurations where some lvols were not mirrored or mirrored to the same disk, this cause problems for sure.

Could you send a vgdisplay -v and the first lines of lvdisplay -v for this VG?

[Robert_Jewell]

I have certainly seen where  a failing disk caused corruption in a filesystem.  If that failing disk has not completely failed and is still attached to the VG, well, then mirroing does its job and copies the corrupt data to the other disk.  

Its certainly strange that this would happen for all of the lvols on that disk though.

Basically, mirroring is not a substitute for data backups.

-Bob

[Torsten.]

In 11.31 you have a new feature called "dying disk detection" or something since some years, so this should not happen anymore.