toby wrote:
> terry433iid@yahoo.com wrote:
> > I'm trying to put together a README for SVM (basic stuff). I did some
> > messing with the metainit() commands to create metadevices....but have
> > no idea how to actually use
> > and test them. I just want to demonstrate how a RAID1 mirror works
> > (e.g. write
> > data to metadevice and then take one underlying disk offline to prove
> > the second disk has the data), ...
>
> I was thinking about something similar this week (fault injection so
> that I could simulate recovery). But I was reasonably satisfied by
> experiments such as
> - taking the submirrors on one disk offline
> - (pulling the disk)
> - replacing the disk later and putting the submirrors online again
> (Similar to a backup scenario. I have two questions for the gurus
> though - why isn't a "lockfs -f" recommended before metaoffline? I
> realise that lockfs on the root filesystem isn't a good idea, but
> doesn't "lockfs -f" just flush without taking a lock? The man page did
> not seem to clarify.)
>
> I also tried a "recovery" sequence, similar to:
> - detach submirrors
> - delete metadbs
> - (pull disk and replace)
> - format/layout slices
> - recreate metadbs
> - metareplace -e (my actual test used metadetach+metaclear before
> format, and metainit here)
> - metattach
>
> I need to redo the recovery test with a brand new disk.
>
> Corrections from SVM savants welcome. I now have to work out how a ZFS
> mirrored pool affects the recovery steps.

Okay, I see that if I want to remove a disk experimentally I can simply
offline that side of the pool's mirror - zpool offline. And if I'm
replacing a failed disk, zpool replace. (Offline failed mirror before
removing it?)

But in my tests, I found that after offline'ing the ZFS mirror, then
yanking the drive, the X2100 hangs at zpool status. Both of the SVM
mirrors were offlined before removing the drive. (Do I also need to
delete the metadbs from that drive before pulling it? I tried that too,
did not help.) Can an expert tell me what I'm missing here?