[Gluster-users] split brain due to conflicting trusted.glusterfs.mdata xattr?

[Alan Orth]

Dear list,

Today I received a "Directory not empty" error while trying to remove a
directory from the FUSE mount of a distribute–replicate volume. Looking in
the directory I found a few files with question marks:

-?????????? ? ?      ?         ?            ? ._Log.out

I checked the volume heal info and there were 0 entries for healing. All
bricks healthy, self-heal daemons up, etc. Looking closer at one of these
files on the backend bricks I found that the file had the same sha256sum
within the replica set, and correctly does not exist in any other sets. The
only issue I could find was that the parent directory on several bricks had
a different trusted.glusterfs.mdata xattr. I removed the directory from
each of the bricks on the other replica sets and then issued `stat` on the
file on the FUSE mount, and it worked. Now the parent directory exists with
the same mdata on each brick.

So my question is: is this a new type of split brain? I don't mind fixing a
few of these manually (especially since I was trying to remove these files
anyways), but it would be good to know more. We are using GlusterFS 8.5 on
CentOS 7.

Thank you!

-- 
Alan Orth
alan.orth at gmail.com
https://picturingjordan.com
https://englishbulgaria.net
https://mjanja.ch
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.gluster.org/pipermail/gluster-users/attachments/20211026/d1ad2bef/attachment.html>