Markus W Kilbinger writes:
> >>>>> "Greg" == Greg Oster <oster@cs.usask.ca> writes:
> 
>     Greg> Well... if you have hot-swap drives, then you could just
>     Greg> stuff the new drive in place of the old one, and not worry
>     Greg> about doing a copyback.. (i.e. the new drive you put in
>     Greg> becomes the hot spare).
>     >> 
>     >> Hmm, how to come!? The old/new drive still has state 'failed',
>     >> the spare is 'used_spare'. So, what will happen if another
>     >> drive fails in this stage?
> 
>     Greg> You stuff another drive into the box, hot-add it, and
>     Greg> rebuild to it :)
> 
> Let's proceed: ;-)
> 
> I just can place in a new drive in place of a failed one. -> A known,
> but 'failed' scsi id / /dev/sd.. refers to a fresh drive now. ->
> Adding this as a new spare ('raidctl -a /dev/sd..', I guess) would
> work!? (I wouldn't even dare...)

Hmm... I don't think I've ever tried exactly what you suggest :) 
I have put a fresh drive (with a SCSI ID different from all other drives, 
including the failed drive) into the system, and hot-added that... 
I'm not sure if it would let you hot-add /dev/sd1e even after /dev/sd1e 
failed, was removed, and was added back in again... (if the device was closed, 
it might, but otherwise it wouldn't...)

>     >> How will the 'failed' drive become the new spare _without_
>     >> reboot?
> 
>     Greg> You stuff a fresh drive into that slot, and when a different
>     Greg> drive fails, rebuild to the fresh drive. Not quite the same
>     Greg> as a spare, but works much the same. :)
> 
> But then you've lost the real spare / redundancy effect, because then
> the first reconstruction step (onto the spare drive), normally done
> automatically, has to be done manually, now.

Except the first reconstruction isn't done automatically right now either :-}
 
> -> Then it would be better to handle the complete spare drives
> manually, because there are only smooth reconstructions. Having a
> 'used_spare' means either blocking copyback or a (reordering) reboot.
> 
> Or is 'raidctl -C ...' allowed with a used running raid!? (Another
> thing I wouldn't dare! ;-))

No.. you don't want to be doing 'raidctl -C ...'s after you get a set 
configured, and it won't let you do 'raidctl -C' when it's already running.
(You really only want to do the 'raidctl -C' the first time, and then let 
the autoconfig stuff do its magic after that :) )

Later...

Greg Oster