Converting a non-HAST ZFS pool to a HAST pool

[Freddie Cash]

On Sat, Oct 16, 2010 at 3:28 PM, Pawel Jakub Dawidek <pjd at freebsd.org> wrote:
> On Fri, Oct 15, 2010 at 11:37:34AM -0700, Freddie Cash wrote:
>> Has anyone looked into, attempted, or considered converting a non-HAST
>> ZFS pool configuration into a HAST one?  While the pool is live and
>> the server is in use.  Would it even be possible?
>>
>> For example, would the following work (in a pool with a single raidz2
>> vdev, where the underlying GEOM provider is glabel)
>>   - zpool offline 1 drive  (pool is now running degraded)
>>   - configure hastd in master mode with a single provider using the
>> "offline" disk (hast metadata takes the place of glabel metadata)
>
> HAST metadata takes much more space than glabel metadata. The latter
> takes only one sector, while the former depends on provider size, but we
> have to keep entire extent bitmap there, so definitely more than one
> sector.

Hrm, darn.  So close.  :)

>>   - zpool replace the offline drive using the /dev/hast/whatever provider
>>   - wait for ZFS to resilver the device
>>   - repeat for each disk in the vdev
>>
>> Is it possible to change the hast configuration and reload hastd to
>> pick up new providers, without restarting it?  Without losing the
>> /dev/hast devices?  Or would it require a restart, which would cause
>> the /dev/hast/* devices to disappear, thus b0rking the pool?
>
> It is possible in 8-STABLE.

Good to know.

>> Just wondering how hard it's going to be to move from our current
>> setup (backup server 1 rsync'd to backup server 2) to a HAST setup
>> (server 1 hast master with server 2 hast slave).
>>
>> If the above won't work, then I guess I'll have to grab a third
>> storage box, configure that as a hast box, copy the data over, destroy
>> server 2, configure it as a hast box, and let them sync.
>
> I'm afraid you have to backup your pool, recreate it on HAST provider
> and recover.

That's what I figured, and why I thought I'd post before attempting
it.  :)  Thanks for the info.

-- 
Freddie Cash
fjwcash at gmail.com