Gluster users - May 2016 - Gluster volume heal info healed shows the healed entries under wrong brick

Hello Gluster developer
I have a gluster cluster with 2 nodes configured as replicated volumes, 
each node has one brick.

I did single node failure test including killing all gluster processes and 
power off one node. 

The node failed is node_f and the other node is node_h. During the 
node_f's failure period, I updated a file 'testMay24' (from a client
machine other than the 2 nodes). I checked the file update is correct 
consumed on the healthy node/brick node_h. 

Then I started the processes OR power on the failure node node_f. After 
that I found the file updated during the node_f failure was healed on 
node_f (with the updated content).

I checked the self-healing information by using 'gluster volume heal 
volume_name info healed'. I found that the healed entries are all listed 
under the node_h, which I supposed should be listed under node_f. Also the 
old healed entries on node_f previous to my test was lost. There are no 
entries that needs heal and no entries of split-brain.

I guess this is related to the self-heal daemon's restart on node_f. Is 
this designed like this or my misunderstanding or a bug? Thanks in advance 
for your help!

Example?
root at node_f:~# gluster volume heal gv0 info healed
Gathering Heal info on volume gv0 has been successful

Brick node_h:/glusterfs/brick1/gv0
Number of entries: 20
at                    path on brick
-----------------------------------
[...]
2016-05-24 06:40:15 /sophytest/testMay24 => This is the file I updated 
during the node_f's failure time. I supposed it should be listed under 
node_f instead of node_h.
[...]

Brick node_f:/glusterfs/brick1/gv0
Number of entries: 0
=>There were some old healed entries that are previous to my test time, 
but were lost after test.


Regards,
Sophy

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://www.gluster.org/pipermail/gluster-users/attachments/20160524/15f8be40/attachment.html>

On 24/05/2016 7:39 PM, Qiu Jie QJ Li wrote:
> checked the self-healing information by using 'gluster volume heal 
> volume_name info healed'. *I found that the healed entries are all 
> listed under the node_h, which I supposed should be listed under 
> node_f. Also the old healed entries on node_f previous to my test was 
> lost.* There are no entries that needs heal and no entries of split-brain.
Heal entries are a bit counter-intuitive - they list the files that have 
been written to and need to be replicated back to the brick that was down.

-- 
Lindsay Mathieson

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://www.gluster.org/pipermail/gluster-users/attachments/20160525/54dd9362/attachment.html>