Description of problem:
Data gain is happening while brick is down and rename a file.
Bringing the brick up and shd is healing this file which leads to creation of the old file on all the bricks. On mount point also we can see the old file with same original size.
Version-Release number of selected component (if applicable):
[root@apandey glu]# gluster --version
glusterfs 3.9dev built on Jul 25 2016 15:09:43
Repository revision: git://git.gluster.com/glusterfs.git
Copyright (c) 2006-2011 Gluster Inc. <http://www.gluster.com>
GlusterFS comes with ABSOLUTELY NO WARRANTY.
Steps to Reproduce:
1. Create a volume (4+2) and mount it.
2. Create a file , file.txt, on mount point.
3. Kill any one brick and rename file.txt to newfile.txt
4. Bring the brick up by force start of volume.
5. List files on mount point. we can see files with old name and new name.
File with old name exist on mount point along with new name.
Only file with new name should exist on mount point. Also heal should happen successfully for the killed brick in which new file frgment should be create and old file should be removed.
Note that this problem happens in glusterfs 3.12.14 as well.
*** Bug 1642638 has been marked as a duplicate of this bug. ***
Still an issue? What should be we doing to fix it?
*** Bug 1698861 has been marked as a duplicate of this bug. ***
Yes, This has not been fixed yet and a design discussion is going on. There is a common approach which will be used for this as well as gfid split brain in afr
Solution could be complex and we are working on all the cases before finalizing design and implementation.
There is one more bug raised by Nithya, https://bugzilla.redhat.com/show_bug.cgi?id=1698861.
I am closing the other bug as duplicate to this.
*** Bug 1757307 has been marked as a duplicate of this bug. ***
This bug is moved to https://github.com/gluster/glusterfs/issues/986, and will be tracked there from now on. Visit GitHub issues URL for further details