Re: Replica self-heal issue (gluster 3.4.2)

Jia-Hao Chen <cjhecm@xxxxxxxxx> · Thu, 17 Apr 2014 17:15:46 +0800

files are present in dir1 and dir2 on node1, and dir1 in node2
dir2 on node2 is empty

On node 1:[/mnt/gluster] # ls /home/harry/gluster/brick/dir*
/home/harry/gluster/brick/dir1:
a  b  c

/home/harry/gluster/brick/dir2:
a  b  c

[/mnt/gluster] # getfattr -m . -d -e hex /home/harry/gluster/brick/dir*
getfattr: Removing leading '/' from absolute path names
# file: home/harry/gluster/brick/dir1
system.posix_acl_access=0x0200000001000700ffffffff020007000000000002000500feff00000200070042420f000200050043420f000200050044420f0004000700ffffffff10000500ffffffff20000500ffffffff
system.posix_acl_default=0x0200000001000700ffffffff020007000000000002000500feff00000200070042420f000200050043420f000200050044420f0004000700ffffffff10000700ffffffff20000000ffffffff
trusted.afr.GV0_DATA-client-0=0x000000000000000000000000
trusted.afr.GV0_DATA-client-1=0x00000000ffffffffffffffff
trusted.gfid=0x28d502fdb1c94ce0ad631d4ab131add0
trusted.glusterfs.dht=0x000000010000000000000000ffffffff

# file: home/harry/gluster/brick/dir2
system.posix_acl_access=0x0200000001000700ffffffff020007000000000002000500feff00000200070042420f000200050043420f000200050044420f0004000700ffffffff10000500ffffffff20000500ffffffff
system.posix_acl_default=0x0200000001000700ffffffff020007000000000002000500feff00000200070042420f000200050043420f000200050044420f0004000700ffffffff10000700ffffffff20000000ffffffff
trusted.afr.GV0_DATA-client-0=0x000000000000000000000000
trusted.afr.GV0_DATA-client-1=0x000000000000000100000001
trusted.gfid=0x28d502fdb1c94ce0ad631d4ab131add0
trusted.glusterfs.dht=0x000000010000000000000000ffffffff

On node 2:
[/mnt/gluster] # ls /home/harry/gluster/brick/dir*

/home/harry/gluster/brick/dir1
a  b  c

/home/harry/gluster/brick/dir2

[/mnt/gluster] # getfattr -m . -d -e hex /home/harry/gluster/brick/dir*
getfattr: Removing leading '/' from absolute path names
# file: home/harry/gluster/brick/dir1
system.posix_acl_access=0x0200000001000700ffffffff020007000000000002000500feff00000200070042420f000200050043420f000200050044420f0004000700ffffffff10000500ffffffff20000500ffffffff

system.posix_acl_default=0x0200000001000700ffffffff020007000000000002000500feff00000200070042420f000200050043420f000200050044420f0004000700ffffffff10000700ffffffff20000000ffffffff
trusted.afr.GV0_DATA-client-0=0x000000000000000000000000
trusted.afr.GV0_DATA-client-1=0x000000000000000000000000
trusted.gfid=0x28d502fdb1c94ce0ad631d4ab131add0
trusted.glusterfs.dht=0x000000010000000000000000ffffffff

# file: home/harry/gluster/brick/dir2
system.posix_acl_access=0x0200000001000700ffffffff020007000000000002000500feff00000200070042420f000200050043420f000200050044420f0004000700ffffffff10000500ffffffff20000000ffffffff
system.posix_acl_default=0x0200000001000700ffffffff020007000000000002000500feff00000200070042420f000200050043420f000200050044420f0004000700ffffffff10000700ffffffff20000000ffffffff

trusted.gfid=0x28d502fdb1c94ce0ad631d4ab131add0

On Thu, Apr 17, 2014 at 11:37 AM, Ravishankar N <ravishankar@xxxxxxxxxx> wrote:

    On 04/16/2014 07:47 PM, Jia-Hao Chen
      wrote:

      Dear all,

        I create a replicated volume with 2 nodes.
        And I make a directory and create a few files in it.

        [/mnt/gluster] # mkdir dir1

          [/mnt/gluster] # echo 123 >> dir1/a

          [/mnt/gluster] # echo 123 >> dir1/b

          [/mnt/gluster] # echo 123 >> dir1/c

        Then I bring down node 2 and rename dir1 to dir2

        [/mnt/gluster] # mv dir1 dir2

        After bring back node 2, dir1 reappear

        [/mnt/gluster] # ls -l

          drwxr-x---    2 harry    harry      4096 Apr 16 11:30
            dir1/
          drwxr-x---    2 harry    harry      4096 Apr 16 11:30
            dir2/

          It surprised me. But it makes sense since dir1 is still
            on node 2.
          When node 2 come back, dir1 is healed on another node.
          However when I list both dir1 and dir2, the file list is
            empty.

          [/mnt/gluster] # ls -l dir1

            [/mnt/gluster] # ls -l dir2

          [/mnt/gluster] #

          It seems a bug to me.
          What is the expected result of this case?

    This is strange. The expected result is that node 2 must also
    contain only dir2. i.e. the self-heal must happen from node 1 to
    node 2 once the latter comes back online.

    Are the files present in the backend bricks?

    -Ravi

          Best regards,
          Chen, Chia-Hao

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

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