bsdlabel blues again

[Volker]

Michel,

On 12/23/-58 20:59, Michel Talon wrote:
> Volker said:
> 
>> As I've done this procedure twice yesterday and once more today,
>> I've double and triple checked everything but I'm running into one
>> single problem:
>>
>> partition c extends past end of unit and doesn't start at 0.
> 
> I think you should run bsdlabel on the mirror, not on the
> raw partitions or disks. Then you will have no more problem 
> of the above type, only the innocuous following one:

Thanks for your answer but I should have stated in my first message
I'm not a totally stupid bloody newby. I do know the difference
between raw disk slices and mirrored slices.

I've also done this procedure a lot times before with less trouble.

I was talking about something what I would like to name a bug or
call it strange behavior.

Just for the archives, I've got it working now.

As I've converted from one (working) mirror to another, gmirror
seemed to cache slice / partition infos. To work around that kind of
trouble, one needs to create the slices, label the mirror, reboot
(!) and then continue with partioning. Otherwise gmirror is working
on old, cached values which will give out of bound partitioning values.

Creating the slices, label the mirror and create the partitions in
one go will most likely get you to bad partition values.

This extra step of rebooting should most likely not be needed if one
is able to `gmirror load' after labeling the mirror but in my case
I've been unable to unload gmirror because I've already been working
on a live mirror. So the reboot was needed as there's no command to
re-sync geom values.

My description does not mean to be technically correct but this is
what I've experienced and the only thing I can imagine what's going
on in the background.

Greetings,

Volker