[Gluster-users] Unable to self-heal contents of '<gfid:00000000-0000-0000-0000-000000000001>'

[Mark Ruys]

Dear Ravi,

You hit the nail right on the head. After successful MD5sum’s I reseted the the afr extended attributes of the bricks as you proposed and self heal daemon now runs without complains.

Thanks,

Mark

On 25 november 2013 at 12:22:27, Ravishankar N (ravishankar at redhat.com) wrote:

On 11/25/2013 01:47 AM, Mark Ruys wrote:
So I decided to bite the bullet and upgraded from 3.3 to 3.4. Somehow this was a painful proces for me (the glusterfs daemon refused to start), so I decided to configure our Gluster pool from scratch. Everything seems to work nicely, except for the self-heal daemon. In the logs, I get every 10 minutes the following line:

[2013-11-24 19:50:34.495204] E [afr-self-heal-common.c:197:afr_sh_print_split_brain_log] 0-GLUSTER-SHARE-replicate-0: Unable to self-heal contents of '<gfid:00000000-0000-0000-0000-000000000001>' (possible split-brain). Please delete the file from all but the preferred subvolume.- Pending matrix:  [ [ 0 2 ] [ 2 0 ] ]

I’ve removed and recreated the .glusterfs/00/00/00000000-0000-0000-0000-000000000001, but that doesn’t seem to make a difference. 

How to fix the self-heal daemon?

Mark

# find . -name 00000000-0000-0000-0000-000000000001 -ls
1447202    0 ----------   2 root     root            0 Nov 23 22:35 ./export-share-1/.glusterfs/indices/xattrop/00000000-0000-0000-0000-000000000001
1319116    0 lrwxrwxrwx   1 root     root            8 Nov 23 22:35 ./export-share-1/.glusterfs/00/00/00000000-0000-0000-0000-000000000001 -> ../../..

Brick 1:
# getfattr -m . -d -e hex export-share-1
# file: export-share-1
trusted.afr.GLUSTER-SHARE-client-0=0x000000000000000000000000
trusted.afr.GLUSTER-SHARE-client-1=0x000000000000000200000000
trusted.gfid=0x00000000000000000000000000000001
trusted.glusterfs.dht=0x000000010000000000000000ffffffff
trusted.glusterfs.quota.dirty=0x3000
trusted.glusterfs.quota.size=0x0000000000000000
trusted.glusterfs.volume-id=0xe6eb05aabe3b456cbf3027275faa529c

Brick 2:
# getfattr -m . -d -e hex export-share-2
# file: export-share-2
trusted.afr.GLUSTER-SHARE-client-0=0x000000000000000200000000
trusted.afr.GLUSTER-SHARE-client-1=0x000000000000000000000000
trusted.gfid=0x00000000000000000000000000000001
trusted.glusterfs.dht=0x000000010000000000000000ffffffff
trusted.glusterfs.quota.dirty=0x3000
trusted.glusterfs.quota.size=0x0000000000000000
trusted.glusterfs.volume-id=0xe6eb05aabe3b456cbf3027275faa529c

From the afr extended attributes, it seems you have hit a metadata-split-brain of the top level (brick) directory (having gfid 01). If you are you able to perform I/O  on all files from the mount point without error (EIO) and the  file contents are identical on both the bricks (check with md5sum), you could  safely clear the afr extended attributes of the bricks:

setfattr -n trusted.afr.GLUSTER-SHARE-client-0  -v 0x000000000000000000000000 /export-share-1
setfattr -n trusted.afr.GLUSTER-SHARE-client-1  -v 0x000000000000000000000000 /export-share-1

setfattr -n trusted.afr.GLUSTER-SHARE-client-0  -v 0x000000000000000000000000 /export-share-2
setfattr -n trusted.afr.GLUSTER-SHARE-client-1  -v 0x000000000000000000000000 /export-share-2

Thanks,
Ravi




_______________________________________________
Gluster-users mailing list
Gluster-users at gluster.org
http://supercolony.gluster.org/mailman/listinfo/gluster-users

---
Oudenhof 4c, 4191NW Geldermalsen, The Netherlands
Web site and travel directions:   www.peercode.nl
Phone +31.88.0084124   ::   Mobile +31.6.51298623
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20131125/0703d583/attachment.html>