On Sun, Dec 27, 2009 at 12:03 AM, Michael Disserman <[email protected]> wrote:
>>Since it's domU, if all you need is increase the rpool disk size,
> there's probably a much simpler method. Extend the disk backing store
> (like by resising the zvol). The hard part then is adjusting the
> partition and slice (which I haven't personally tried) so that zfs
> will recognize the new disk size.
>
> that's one of the ways however I don't know how correctly extend the 
> partition and how to extend ZFS when the new partition table is set. There's 
> no any resizezfs like i.e. resize2fs in Linux.

Ah, OK, now we're getting somewhere :D

What happens is that zfs (until now, I think) will simply use all
available space. So it automatcially extends itself to occupy the
available space after a boot or import. There's no need for a
resize2fs-equivalent. The term "available space" can differ in meaning
though.

When used directly on top of a block device (like on top of LVM with
zfs-fuse on Linux), if you extend that LV and do export-import of the
pool, zfs will recognize the new size (this is what I usually do).
Same thing goes with partition, if you put zfs on partiton, extend the
partition (i.e. with parted or similar) and zfs will recognize the new
size after the next export-import.

In rpool, zfs uses slice. I don't now of any program that can safely
extend a solaris slice, so you MIGHT be able to achieve a similar
effect by writing down the old partition table and solaris vtoc, then
creating a new one that reflects the new size. Note that I haven't
tried this personally (never had the need to), but this was mentioned
some time ago in this list.

As I recall there were also plans to make zfs be able to recoginze the
new size without export/import, and that the "automatic extend"
feature should be tunable via a zpool property, but I haven't heard
about it for some time, so most likely they still behave like what I
mentioned above.

-- 
Fajar
_______________________________________________
xen-discuss mailing list
[email protected]

Reply via email to