Take 4 servers on which the exports are structured as:

/testdirs/4d-3r-master/disk1
/testdirs/4d-3r-master/disk2
/testdirs/4d-3r-master/disk3

Together these will form 12 bricks, (4 servers X 3 bricks each)

On qa26, to start a distributed-replicated with 3 replicas, I run:

# gluster volume create 4dist-3repl  replica 3 192.168.1.78:/testdirs/4d-3r-master/disk1/ 192.168.1.78:/testdirs/4d-3r-master/disk2/ 192.168.1.78:/testdirs/4d-3r-master/disk3/ 192.168.1.79:/testdirs/4d-3r-master/disk1/ 192.168.1.79:/testdirs/4d-3r-master/disk2/ 192.168.1.79:/testdirs/4d-3r-master/disk3/ 192.168.1.80:/testdirs/4d-3r-master/disk1/ 192.168.1.80:/testdirs/4d-3r-master/disk2/ 192.168.1.80:/testdirs/4d-3r-master/disk3/ 192.168.1.77:/testdirs/4d-3r-master/disk1/ 192.168.1.77:/testdirs/4d-3r-master/disk2/ 192.168.1.77:/testdirs/4d-3r-master/disk3/

After starting the volume file has the first replicate generated as:

volume dr-client-0
    type protocol/client
    option transport-type tcp
    option remote-host 192.168.1.77
    option transport.socket.nodelay on
    option remote-subvolume /testdirs/4d-3r-master/disk1
end-volume

volume dr-client-1
    type protocol/client
    option transport-type tcp
    option remote-host 192.168.1.78
    option transport.socket.nodelay on
    option remote-subvolume /testdirs/4d-3r-master/disk1
end-volume

volume dr-client-2
    type protocol/client
    option transport-type tcp
    option remote-host 192.168.1.79
    option transport.socket.nodelay on
    option remote-subvolume /testdirs/4d-3r-master/disk1
end-volume

volume dr-client-3
    type protocol/client
    option transport-type tcp
    option remote-host 192.168.1.80
    option transport.socket.nodelay on
    option remote-subvolume /testdirs/4d-3r-master/disk1
end-volume

volume dr-replicate-0
    type cluster/replicate
#   option read-subvolume on
#   option favorite-child on
#   option background-self-heal-count on
#   option data-self-heal on
#   option data-self-heal-algorithm on
#   option data-self-heal-window-size on
#   option metadata-self-heal on
#   option entry-self-heal on
#   option data-change-log on
#   option metadata-change-log on
#   option entry-change-log on
#   option strict-readdir on
    subvolumes dr-client-0 dr-client-1
end-volume

See attached file for complete vol file. I think to reproduce, we will not need 4 servers, just three bricks on single server may suffice to show the above behaviour.

Worked for me on a single system.. will try over more machine, and see if the behavior is different..

checked on 4 machines. It worked fine.

i checked it again with 12 bricks and it worked this time. Will re-open if i see it again. thanks.