On 10/29/2011 11:43 AM, Emmanuel Dreyfus wrote:
Emmanuel Dreyfus<manu@xxxxxxxxxx> wrote:
So that means stat inode is afr_itransform'ed but not readdir inode,
which is be the opposite of what I said earlier.
That beahviour is fixed by making sure readdir and readdirp get the same
afr_itransform treatment.
But doing so means restoring a afr_itransform() that was removed by the
patch backporting the inode-generated-from-gfid feature from HEAD. So I
suspect I do it the wrong way around, and instead of adding
afr_itransform() to readdirp_cbk(), I should remove it evrywhere it
remains in afr code, that is from afr_fresh_lookup_cbk() and
afr_revalidate_lookup_cbk().
What is the goal? Should afr_itransform() be completely killed?
Yes. That is right. Also dht_itransform should go. To achieve that in
release-3.2 branch, below patches should go in.
http://review.gluster.com/632 <-- for replicate itransform removing
http://review.gluster.com/613 <--- for distribute itransform removing.
Sorry for delaying answering it in one shot. most of us were away as it
was a festive season here in India. (and one of my replies was from
mobile device, so couldn't enter all these).
Regards,
Amar
