Re: [Gluster-devel] [PATCH] Enable replica 1 mirror creation for mirrored volumes

[Joe Landman]

On 05/29/2011 07:31 PM, Craig Carl, Gluster wrote:
> Avati - This type of functionality is incredibly useful from an IT
> operations POV, my thinking is it would make migrating from a
> distribute only to a distribute-replica volume easier.

Actually, this is what I had in mind.  This was only the very first 
step.  I was starting to look at the logic underneath the replicate 
setup.  My initial guess is that all it would really take to enable this 
is a sequence of

        create with missing mirror
        add-brick to mirror

Then growth (add-brick) to grow.  Rebalance after add.

I don't think there needs to be much additional code for this.  Will be 
trying it soon on one of our internal units.  The change in this case is 
a single line really, and it just lowers the limit for the mirror to 1.

An alternative, and similar to linux MD RAID, is to use the "missing" 
keyword. This could be useful in setting up a larger mirror with initial 
portion missing.  This might make sense for geo-replication as well 
(create 1/2 of the replica, and add the next portion as a mirror, 
leveraging rsync and inotify events).



> Craig
>
>
> ----- Original Message ----- From: "Anand
> Avati"<address@hidden> To: address@hidden Cc:
> "Gluster Devel"<address@hidden> Sent: Saturday, May 28,
> 2011 8:12:18 AM Subject: Re: [Gluster-devel] [PATCH] Enable replica 1
> mirror creation for mirrored volumes
>
>
> Joe, Can you shed some more light on the use case? What purpose is
> the replicate translator serving without the second server? In what
> way is this more beneficial than loading the replicate translator
> only when the second server is introduced to the system?
>
>
> Avati
>
>
> On Sat, May 28, 2011 at 7:11 PM, Joe Landman<
> address@hidden>  wrote:
>
>
> This patch to the cli enables a very simple replica 1 mirror creation
> (similar to the "missing" attribute for mdadm RAID1 creation).
>
> It will likely need additional supporting work in add-brick so that
> incomplete mirrors are assembled correctly for distributed-replicated
> cases.
>
> Building/Assembling an incomplete mirror is an important capability
> needed for a number of use cases.


--
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics, Inc.
email: address@hidden
web  : http://scalableinformatics.com
       http://scalableinformatics.com/sicluster
phone: +1 734 786 8423 x121
fax  : +1 866 888 3112
cell : +1 734 612 4615