zfs discuss - Sep 2009 - Adding new disks and ditto block behaviour

I have machine that had 2x 1TB drives in it. They were in the same zpool and
that entire zpool is set to "copies=2". From what I understand this
will store all my data twice, and if the SPA is doing its job right it will
store the copies on different disks and store the checksum for any given block
on the other disk.

I filled this with data. So I added a 1.5 TB drive to the pool. Where will my
ditto blocks and checksums go? Will it migrate data from the other drives
automatically? Will it migrate data if I scrub or re-silver? will it never
migrate data and just store all the new blocks and checksums on the new drive?

I am interested in making sure that a single drive failure will not cause loss
of data, what should I do if anything?
-- 
This message posted from opensolaris.org

Joe Toppi wrote:
> I have machine that had 2x 1TB drives in it. They were in the same zpool
and
> that entire zpool is set to "copies=2". From what I understand
this will
> store all my data twice, and if the SPA is doing its job right it will
store
> the copies on different disks and store the checksum for any given block on
> the other disk.
> 
> I filled this with data. So I added a 1.5 TB drive to the pool. Where will
my
> ditto blocks and checksums go? Will it migrate data from the other drives
> automatically? Will it migrate data if I scrub or re-silver? will it never
> migrate data and just store all the new blocks and checksums on the new
> drive?
> 
> I am interested in making sure that a single drive failure will not cause
> loss of data, what should I do if anything?
Use mirrors, raidz1, or raidz2. copies=2 will _not_ help you in case of drive 
failure, as the pool will refuse to import.

-- 
Carson

On Thu, 17 Sep 2009, Joe Toppi wrote:
> I filled this with data. So I added a 1.5 TB drive to the pool. 
> Where will my ditto blocks and checksums go? Will it migrate data 
> from the other drives automatically? Will it migrate data if I scrub 
> or re-silver? will it never migrate data and just store all the new 
> blocks and checksums on the new drive?
Zfs does not automatically migrate data just because you added more 
drives.  Scrub will only migrate failing data blocks.  Resilver 
clones a failing disk.  When a vdev becomes very full, more writes 
will be directed to the empty devices.

If you have enough free disk space to store everything you had before, 
plus lots of space to spare, you could try creating a new filesystem 
in the pool and using zfs send to send from the existing filesystem to 
the new filesystem, and then destroy the old filesystem once you are 
satisified with the new one.  This would only work if there is 
considerably more free space than existing data and the result will 
still be lop-sided.

If you have a whole lot of reliable storage space elsewhere, you could 
use zfs send to send to a file in that other storage space, destroy 
the old filesystem, and then recreate it with your zfs send file.

Bob
--
Bob Friesenhahn
bfriesen at simple.dallas.tx.us, http://www.simplesystems.org/users/bfriesen/
GraphicsMagick Maintainer,    http://www.GraphicsMagick.org/

Yeah, after I learned that ditto blocks don''t protect against failed
drives, I started working on a plan to move to raidz. I couldn''t find
any good documentation on setting multiple filesystems systems in one
pool, though I know it is possible. I think I have enough storage to
work this together somehow, but I think I need to read more and plan
more.

Thanks for your response

On Thu, Sep 17, 2009 at 8:12 PM, Bob Friesenhahn
<bfriesen at simple.dallas.tx.us> wrote:
> On Thu, 17 Sep 2009, Joe Toppi wrote:
>
>> I filled this with data. So I added a 1.5 TB drive to the pool. Where
will
>> my ditto blocks and checksums go? Will it migrate data from the other
drives
>> automatically? Will it migrate data if I scrub or re-silver? will it
never
>> migrate data and just store all the new blocks and checksums on the new
>> drive?
>
> Zfs does not automatically migrate data just because you added more drives.
> ?Scrub will only migrate failing data blocks. ?Resilver clones a failing
> disk. ?When a vdev becomes very full, more writes will be directed to the
> empty devices.
>
> If you have enough free disk space to store everything you had before, plus
> lots of space to spare, you could try creating a new filesystem in the pool
> and using zfs send to send from the existing filesystem to the new
> filesystem, and then destroy the old filesystem once you are satisified
with
> the new one. ?This would only work if there is considerably more free space
> than existing data and the result will still be lop-sided.
>
> If you have a whole lot of reliable storage space elsewhere, you could use
> zfs send to send to a file in that other storage space, destroy the old
> filesystem, and then recreate it with your zfs send file.
>
> Bob
> --
> Bob Friesenhahn
> bfriesen at simple.dallas.tx.us,
http://www.simplesystems.org/users/bfriesen/
> GraphicsMagick Maintainer, ? ?http://www.GraphicsMagick.org/
>


-- 
- Joe Toppi
(402) 714-7539
toppij at gmail.com
http://www.assuredts.com/toppij/