[Gluster-devel] inodes and AFR

[Anand Avati]

On Wed, Nov 30, 2011 at 10:34 AM, Emmanuel Dreyfus <manu at netbsd.org> wrote:
> Anand Avati <anand.avati at gmail.com> wrote:
>
>> Yes, that was the very intention of making the inode number a
>> derivation of the GFID instead of the backend inode number, so that
>> brick addition/replacement does not impact its stability.
>
> I tested it and it does not work as expected on 3.2.5. add-brick and
> rebalance migrate-data do change inode numbers.
>
> Inode numbers, at the beginnig
> /pfs    1
> /pfs/manu       12472916913858460615
> /pfs/manu/openssl-0.9.8o.tar    12709833430099732973
>
> After add-brick, everything changed:
> /pfs    1
> /pfs/manu       6499089754007369614
> /pfs/manu/openssl-0.9.8o.tar    6972922786489914330
>
> After rebalance fix-layout (nothing changed)
> /pfs    1
> /pfs/manu       6499089754007369614
> /pfs/manu/openssl-0.9.8o.tar    6972922786489914330
>
> After rebalance migrate-data (file changed but not directory)
> /pfs    1
> /pfs/manu       6499089754007369614
> /pfs/manu/openssl-0.9.8o.tar    4597243647138552351

The backporting of the gfid-derived inode number is incomplete in 3.2
as the itransforms are not NOP'ed. We will fix this.

Avati