[Gluster-devel] Simplify creation and set-up of meta-volume (shared storage)

[Shubhendu Tripathi]

Avra,

I understand its not planned for Everglades :)

Also we are planning to provide mounting of meta volume on the nodes of 
the cluster from Console as part of volume syncing and addition of new 
nodes to the cluster. Looks like if this option is set, the explicit 
mounting of the meta volume from console is not required and it would be 
taken care by gluster.

Currently we show explicit warning during stop/remove-brick of the meta 
volume in console and allow the operations. I dont feel there would be 
any impact due to new feature.

Kindly let us know if any other impact on console or we need to take 
care of anything else as result of this feature.

Thanks and Regards,
Shubhendu

On 05/15/2015 07:30 PM, Avra Sengupta wrote:
> Hi,
>
> A shared storage meta-volume is currently being used by 
> snapshot-scheduler, geo-replication, and nfs-ganesha. In order to 
> simplify the creation and set-up of the same, we are introducing a 
> global volume set option(cluster.enable-shared-storage).
>
> On enabling this option, the system analyzes the number of peers in 
> the cluster, which are currently connected, and chooses three such 
> peers(including the node the command is issued from). From these peers 
> a volume(gluster_shared_storage) is created. Depending on the number 
> of peers available the volume is either a replica 3 volume(if there 
> are 3 connected peers), a replica 2 volume(if there are 2 connected 
> peers), or a single brick volume(if there is only one node in the 
> cluster). "/var/run/gluster/ss_brick" serves as the brick path on each 
> node for the shared storage volume. We also mount the shared storage 
> at "/var/run/gluster/shared_storage" on all the nodes in the cluster 
> as part of enabling this option.
>
> Once the volume is created, and mounted the maintainance of the volume 
> like adding-bricks, removing bricks etc., is expected to be the onus 
> of the user.
>
> On disabling the option, we provide the user a warning, and on 
> affirmation from the user we stop the shared storage volume, and 
> unmount it from all the nodes in the cluster.
>
> These are achieved with hook-scripts as part of the volume set option. 
> If anyone is interested in having a look at the patch, it's available 
> for review at http://review.gluster.org/#/c/10793/ . If there is any 
> feedback and suggestion regarding the same, please feel free to share it.
>
> Regards,
> Avra
>
> _______________________________________________
> Gluster-devel mailing list
> Gluster-devel at gluster.org
> http://www.gluster.org/mailman/listinfo/gluster-devel