Gluster users - Feb 2014 - Question on replicated volumes with bricks on the same server

Hi all,

I would like to know how gluster distribute the data when two bricks
of the same volumes are on the same server. Specifically, I would like
to know if there is any way to spread the replicas on different nodes
whenever possible, in order not to lose any data if the node goes
down.

I did a simple test and it seems that the way replicas are spread over
the bricks is related to the way the volume is created, that is if I
create a volume with:

    gluster volume create vol1 replica 2\
        gluster-data001:/srv/gluster/vol1.1 \
        gluster-data001:/srv/gluster/vol1.2 \
        gluster-data002:/srv/gluster/vol1.1 \
        gluster-data002:/srv/gluster/vol1.2

replicas of a file will be stored on the two bricks of the same
server, while if I create the volume with

    gluster volume create vol1 replica 2\
        gluster-data001:/srv/gluster/vol1.1 \
        gluster-data002:/srv/gluster/vol1.1 \
        gluster-data001:/srv/gluster/vol1.2 \
        gluster-data002:/srv/gluster/vol1.2

replicas will be saved on two bricks of different servers.

So, my guess is that if I create a "replica N" replicated+distributed
volumes using the bricks:

  gluster-1:/srv/gluster
  ...
  gluster-[N*M]:/srv/gluster

gluster internally creates a distributed volumes made of the following
replicated "volumes":

  replicated volume 1: gluster-[1..N]:/srv/gluster
  replicated volume 2: gluster-[N+1..2N]:/srv/gluster
  ...
  replicated volume M: gluster-[N*(M-1)+1..N*M]:/srv/gluster

Is that correct or there is a more complex algorithm involved?

.a.

-- 
antonio.s.messina at gmail.com
antonio.messina at uzh.ch                     +41 (0)44 635 42 22
GC3: Grid Computing Competence Center      http://www.gc3.uzh.ch/
University of Zurich
Winterthurerstrasse 190
CH-8057 Zurich Switzerland

On 02/11/2014 03:21 PM, Antonio Messina wrote:
> Hi all,
>
> I would like to know how gluster distribute the data when two bricks
> of the same volumes are on the same server. Specifically, I would like
> to know if there is any way to spread the replicas on different nodes
> whenever possible, in order not to lose any data if the node goes
> down.
>
> I did a simple test and it seems that the way replicas are spread over
> the bricks is related to the way the volume is created, that is if I
> create a volume with:
>
>      gluster volume create vol1 replica 2\
>          gluster-data001:/srv/gluster/vol1.1 \
>          gluster-data001:/srv/gluster/vol1.2 \
>          gluster-data002:/srv/gluster/vol1.1 \
>          gluster-data002:/srv/gluster/vol1.2
>
> replicas of a file will be stored on the two bricks of the same
> server, while if I create the volume with
>
>      gluster volume create vol1 replica 2\
>          gluster-data001:/srv/gluster/vol1.1 \
>          gluster-data002:/srv/gluster/vol1.1 \
>          gluster-data001:/srv/gluster/vol1.2 \
>          gluster-data002:/srv/gluster/vol1.2
>
> replicas will be saved on two bricks of different servers.
>
> So, my guess is that if I create a "replica N"
replicated+distributed
> volumes using the bricks:
>
>    gluster-1:/srv/gluster
>    ...
>    gluster-[N*M]:/srv/gluster
>
> gluster internally creates a distributed volumes made of the following
> replicated "volumes":
>
>    replicated volume 1: gluster-[1..N]:/srv/gluster
>    replicated volume 2: gluster-[N+1..2N]:/srv/gluster
>    ...
>    replicated volume M: gluster-[N*(M-1)+1..N*M]:/srv/gluster
>
> Is that correct or there is a more complex algorithm involved?
>
The interpretation is correct. The way replica sets are chosen is 
related to the order in which bricks are defined at the time of volume 
creation.

-Vijay