Bug 1145681

Description of problem:
I am in a 3 hosts/3 bricks cluster configuration (Debian), with a single distributed volume.

Because I would like to change the FS of my bricks from xfs to ext4 (xfs is to slow with small files) I performed the following commands :
- remove a brick
- commit the removal
- add the new brick to the pool
- launch rebalancing

The rebalancing finished successfully with some skipped files :
http://pastebin.com/wekv3CRv

The problem is that every skipped files are present at two several location :
- on its source brick : the file is consistent
- on its destination brick : the file is corrupted

Here is an example :
# ls -lrt /glusterfs/projets-brick1/some_folder/file.dat 
-rw-rw----+ 2 some_user some_group 32302 Sep  5 10:34 /glusterfs/projets-brick1/some_folder/file.dat 

# ls -lrt /glusterfs/projets-brick2/some_folder/file.dat 
-rw-rw---T+ 2 some_user some_group 32302 Sep 19 10:01 /glusterfs/projets-brick2/some_folder/file.dat

The difference between the two files is the modification date. One has the good date (in this case brick1) unlike the other file  (and corrupted) has the rebalancing date (in this case brick2).

On the client (direct glusterfs mount or glusterfs over CIFS) each time we refresh the file, the modification date changed. I concluded that one time it was pointing to a brick and the second time to the other brick.

I need to perform these steps twice again, in order to change all my brick's FS type. I would like to be more serene before to continue.

Thx.
T-