"Greg A. Woods" writes:
> [ On Sunday, October 19, 2003 at 15:13:37 (-0600), Greg Oster wrote: ]
> > Subject: Re: another RAIDframe oddity.... 
> >
> > "Greg A. Woods" writes:
> > > BTW, is there any easy way to change the "Last configured as" value in
> > > the component labels? 
> > 
> > Not easy.. no.  (as in, no way to do it directly from raidctl)
> 
> Hmmm.... another issue arose yesterday where this would also come into
> play, and where it's a lot less "cosmetic" than my initial situation.
> 
> I had to move a RAID set from one host to another.  Now since the drives
> were hot-swap I was able to move them without rebooting.  However once I
> had them attached to the new system I had no way to autoconfigure them.
> Simply triggering autoconfiguration again, if there were a way to do it,
> would have caused a "collision" with the existing RAID set's device.
> 
> I think it makes sense to consider supporting hot swap of whole RAID
> sets between systems.

If I can ever toss the old configuration code, the new configuration 
code would do the trick -- just tell it what components are involved, 
and it would figure out how they fit together.  Things like "where 
last configured" could be easily changed.

>  I think this means that besides having some way
> to re-trigger the RAIDframe autoconfiguration procedures there also
> needs to be some way to manually inspect the on-disk component labels of
> arbitrary specified disks, and there needs to be some way to adjust the
> "last configured as" name in those labels so that a "new" RAID set can
> be brought safely on-line without a reboot.  Of course already
> configured devices must be ignored when autoconfiguration runs too.
> There may be other issues, but that's what comes to mind initially.

As above, if the autoconfig code can be used for regular 
configuration (i.e. 'raidctl -c' and 'raidctl -C') then this will 
work.  I had a stab at changing it a couple years ago, but the 
various configuration bits are fiendishly interwoven, and successfully 
removing the old config bits without killing the patient appeared
too difficult at the time...

Later...

Greg Oster