Description of problem:
========================
On a 2x2 dis-rep volume, nodes were replaced with new nodes. After replacing the nodes, split-brain were observed on the "/". 

Version-Release number of selected component (if applicable):
============================================================
glusterfs-3.7.1-10.el6rhs.x86_64

How reproducible:
================
2/2

Steps to Reproduce:
======================
1. Create 2 x 2 dis-rep volume, Start the volume. Create fuse mount

2. From fuse mount execute "self_heal_all_file_types.sh <mount-point> "glusterfs" "create".

3. Bring down any combination randomly from set :
volume_2_2_brick_takedown_combinations = [
    ["Brick1"],
    ["Brick2"],
    ["Brick3"],
    ["Brick4"],
    ["Brick1", "Brick3"],
    ["Brick1", "Brick4"],
    ["Brick2", "Brick3"],
    ["Brick2", "Brick4"], ]

4. Wait for the IO to complete and calculate arequal-checkm 

5. Perform replace-brick on the offline brick. 

6. Wait for the self-heal to complete. 

7. Verify arequal-checksum after self-heal and before replace-brick. (Both should match)

8. From fuse mount execute "self_heal_all_file_types.sh <mount-point> "glusterfs" "modify".

9. Bring down any combination randomly from set :
volume_2_2_brick_takedown_combinations = [
    ["Brick1"],
    ["Brick2"],
    ["Brick3"],
    ["Brick4"],
    ["Brick1", "Brick3"],
    ["Brick1", "Brick4"],
    ["Brick2", "Brick3"],
    ["Brick2", "Brick4"], ]

10. Wait for the IO to complete and calculate arequal-checkm 

11. Perform replace-brick on the offline brick. 

12. Wait for the self-heal to complete. 

13. Verify arequal-checksum after self-heal and before replace-brick. (Both should match)

Repeat the above until all the combinations are executed. Also , ONce brought down brick combination not to be brought down again. 

Actual results:
===============
While executing the above test case for the 4th time when the bricks were brought offline, there was GFID mismatch in the files. 

Expected results:
================
heal should be successful