[Gluster-users] Possible split brain situation?

[Andrea Veri]

Hi,

while verifying the heal information for one of the hosted bricks I found
out a few entries were not properly in sync anymore (I wondered how that
could be possible as just one node is serving the data and the other is
there as a backup and no, I had no downtime that could have made the
gluster fuse client to switch to the other available node), after following
the documentation at [1] I removed both the offending files and the
respective hard links from the .glusterfs directory, (I found the
'-samefile' flag of 'find' very handy for this job) but then I still had a
few standing entries that would not report what file they were affecting
but just the relevant gfid, in particular:

sudo gluster volume heal mailman info
Brick brick01:/mnt/gluster_volumes/mailman/
<gfid:7220e73e-d0c6-4f03-a541-b59fa713db7b>
<gfid:a58e2209-0281-46e6-91d7-44e3fb06d6fa>
<gfid:c5bd1fe9-a19a-4b32-a5a5-83f9f6d5d425>
<gfid:5985ae98-78c0-4b84-840a-7149093e79dc>
Number of entries: 4

Brick brick02:/mnt/gluster_volumes/mailman/
7220e73e-d0c6-4f03-a541-b59fa713db7b
Number of entries: 1

The relevant log entries:

[2014-08-17 21:04:07.747654] E
[afr-self-heal-common.c:233:afr_sh_print_split_brain_log]
0-mailman-replicate-0: Unable to self-heal contents of
'7220e73e-d0c6-4f03-a541-b59fa713db7b' (possible split-brain). Please
delete the file from all but the preferred subvolume.- Pending matrix:  [ [
0 4 ] [ 2 0 ] ]

Am I in presence of another split brain situation? additionally the process
that should be accessing the files I touched is reporting me an
"Input/Output" error when trying to write to the glusterfs-mounted
directory. (not on the whole brick but just on the directory that had the
original split-brain)

Thanks in advance!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20140817/de378566/attachment.html>