Description of problem:
When the cluster.subvols-per-directory is set to 1 and rebalance is issued after adding one more brick to a 2 node distribute volume, storage domain is going down.


Version-Release number of selected component (if applicable):
glusterfs 3.3.0rhsvirt1


Steps to Reproduce:
1. Create a 2 node distribute volume. Tag the volume with group virt and also set the cluster.subvols-per-directory to 1
gluster v set <volname> cluster.subvols-per-directory 1
2. Us this as a storage domain for RHEV and create some vms on the storage domain. 
3. Now add one more brick to the volume and start rebalance.
  
Actual results:
The storage domain goes down when the rebalance is issued.

Expected results:
Storage domain should not go down.

Additional info:

From 1st brick
[root@rhs-client37 bricks]# getfattr -m . -d -e hex /brick2/brick1/
getfattr: Removing leading '/' from absolute path names
# file: brick2/brick1/
trusted.gfid=0x00000000000000000000000000000001
trusted.glusterfs.dht=0x000000010000000000000000ffffffff
trusted.glusterfs.volume-id=0xcee7cd2ab12a46a0a4549f4c5cd02957

From second brick
[root@rhs-client43 bricks]# getfattr -m . -d -e hex /brick2/brick2
getfattr: Removing leading '/' from absolute path names
# file: brick2/brick2
trusted.gfid=0x00000000000000000000000000000001
trusted.glusterfs.dht=0x000000010000000000000000ffffffff
trusted.glusterfs.volume-id=0xcee7cd2ab12a46a0a4549f4c5cd02957