[Gluster-devel] inodes

[Emmanuel Dreyfus]

Amar Tumballi <amar at gluster.com> wrote:

> > Or is the whole change  self-contained (I mean not dependent 
> > on other changes) and will only bring benefits?
>
> It will surely bring only benefits. 

But it is self contained, right? I do not need to pull other changes?

To give a more precise idea of my motivations: NetBSD FUSE once relied
on kernel name cache. This caused issues with distributed filesystems
such as glusterfs because one client would not see the changes from
another client (for instance, moved files could be refered by their
former names). 

I attempted to fix that and honour FUSE entry ttl (entry_valid in struct
fuse_entry_our), but ran into issues with getcwd(3) getting broken
because LOOKUP, GETATTR and READDIR would not return the same values for
inodes. 

Thank to your hint, I was able to address that by pulling two changes
sets. The first is from this change: 
https://github.com/gluster/glusterfs/commit/333bd9b59c4298a7cb7796f3bd4e
e444376c68c6

I had to significally alter it to get it working. Blended with other
changes I already submitted (linkat, mknod), it turns into the patch
below. It may be intersting to have a look at it. There is a trivial fix
with the uint64_t cast, but there is another fix with READDIRP that I
suspect to be missing in you change.
http://cvsweb.netbsd.org/bsdweb.cgi/pkgsrc/filesystems/glusterfs/patches
/patch-br?rev=1.8

The second patch set is from http://review.gluster.com/632
I had to nuke more afr_itransform than in the patch at
review.gluster.com, which is probably also worth a review. It turns into
this:
http://cvsweb.netbsd.org/bsdweb.cgi/pkgsrc/filesystems/glusterfs/patches
/patch-by?rev=1.1

Now the question is: It seems to work, do I need anything more to
acheive my goal? http://review.gluster.com/613 suggest inode changes
broke rebalance, and that this change is required to restore it. Is it
the way it should be understood?

I also understand you are poorly connected right now. There is no hurry,
I can wait for an answer until you to get back to your desk.

-- 
Emmanuel Dreyfus
http://hcpnet.free.fr/pubz
manu at netbsd.org