Fast RAID rebuild?

Fast RAID rebuild?

[danci]

Hello!

I have a 3 disk RAID-5 array with one spare disk on two IDE controller. As
I'm experimenting with kernel and drivers, sometimes my kernel doesn't
detect one of the controllers and therefore marks one of the RAID disks as
faulty. The RAID-5 is my root fs...

Next time I boot the kernel with all disks detected, RAID starts
rebuilding which takes a long time - pretty unconvenient for testing...

Is there a way to quickly add the disk back to array without resyncing
- something like 'mdadm --unfail'? :)


  D.

Re: Fast RAID rebuild?

[Warren Young]

danci@agenda.si wrote:
> 
> Is there a way to quickly add the disk back to array without resyncing
> - something like 'mdadm --unfail'? :)

md is right not to trust the state of the replaced disk.  Just in order 
to get to the mdadm point, you have to boot up, which means files have 
changed on the live disks relative to the one that was temporarily 
missing.  You'd risk massive corruption by forcing it to accept the new 
disk without rebuilding the array.

You maybe could get away with this in a hardware RAID system because you 
can administer the array before the OS comes up.  None of the RAID cards 
I've ever played with allow this, though.

The real solution to your problem is to put all the disks on the same 
kind of controller.  A 4-port 3Ware card, for instance.  Then all disks 
come up together or none of them do.  You can still use Linux soft RAID 
with most hardware RAID cards, if you must.  Or, if you want to cheap 
out, get a 2-port Promise IDE card and put the three disks on it.

Re: Fast RAID rebuild?

[danci]

On Thu, 24 Jul 2003, Warren Young wrote:

> danci@agenda.si wrote:
> >
> > Is there a way to quickly add the disk back to array without resyncing
> > - something like 'mdadm --unfail'? :)
>
> md is right not to trust the state of the replaced disk.  Just in order
> to get to the mdadm point, you have to boot up, which means files have
> changed on the live disks relative to the one that was temporarily
> missing.  You'd risk massive corruption by forcing it to accept the new
> disk without rebuilding the array.

I thought it shouldn't work and I said so to the guy asking me... But I
just asked to see if *maybe* it is possible somehow.

Thanks for the answer,

   D.