raid1 faulty disk --> hot adding

raid1 faulty disk --> hot adding

[Raz Ben-Jehuda(caro)]

Hello guys.
I tried to to raidhotadd a faulty disk to a mirror and it failed
saying that "bd_claim failed on..."
i took a dive into the  code and it seems that when a disk in the raid
fails the bd_holder fileld in
the block_device struct remained with the old value although the disk
was faulty.
So the logical thing to do was to set rdev->bdev->bd_holder to 0x00 .
I did it in the error() routine in raid1.c .
This way I manage to hotadd a disk.
Are there anyone here that can point his view regarding this matter  ?
--
Raz
Long Live the Penguin

Re: raid1 faulty disk --> hot adding

[Neil Brown]

On Friday June 10, raziebe@gmail.com wrote:
> Hello guys.
> I tried to to raidhotadd a faulty disk to a mirror and it failed
> saying that "bd_claim failed on..."
> i took a dive into the  code and it seems that when a disk in the raid
> fails the bd_holder fileld in
> the block_device struct remained with the old value although the disk
> was faulty.
> So the logical thing to do was to set rdev->bdev->bd_holder to 0x00 .
> I did it in the error() routine in raid1.c .
> This way I manage to hotadd a disk.
> Are there anyone here that can point his view regarding this matter  ?

If a drive fails, and you want to re-add it, what you should do is
first remove it, and then add it.
  raidhotremove /dev/mdX /dev/sdY
  raidhotadd /dev/mdX /dev/sdY

(or equivalent mdadm commands).

clearing bd_holder is certainly not the right thing to do.

NeilBrown

Re: raid1 faulty disk --> hot adding

[Raz Ben-Jehuda(caro)]

Yes. my mistake. 
I tried it with two ramdisks and I made some stupid mistake.

thanks Neil

6/10/05, Neil Brown <neilb@cse.unsw.edu.au> wrote:
> On Friday June 10, raziebe@gmail.com wrote:
> > Hello guys.
> > I tried to to raidhotadd a faulty disk to a mirror and it failed
> > saying that "bd_claim failed on..."
> > i took a dive into the  code and it seems that when a disk in the raid
> > fails the bd_holder fileld in
> > the block_device struct remained with the old value although the disk
> > was faulty.
> > So the logical thing to do was to set rdev->bdev->bd_holder to 0x00 .
> > I did it in the error() routine in raid1.c .
> > This way I manage to hotadd a disk.
> > Are there anyone here that can point his view regarding this matter  ?
> 
> If a drive fails, and you want to re-add it, what you should do is
> first remove it, and then add it.
>   raidhotremove /dev/mdX /dev/sdY
>   raidhotadd /dev/mdX /dev/sdY
> 
> (or equivalent mdadm commands).
> 
> clearing bd_holder is certainly not the right thing to do.
> 
> NeilBrown
> 
> 


-- 
Raz
Long Live the Penguin