Hi Ceph developers, With this email I would like to start a discussion about the design of the consistency groups feature. I created a wiki page for this blueprint: http://tracker.ceph.com/projects/ceph/wiki/consistency_groups So far it's only filled with what I know for sure about the consistency groups feature. First of all this feature is going to require a new entity in rbd - consistency group. This entity will be represented in the object storage in the way similar to how it's done for images. cg_id.<group name> object has the id of the consistency group. cg_header.<group id> object has the list of all images in the consistency group. Snapshot information will be stored in the omap attributes of cg_header. Every description of a snapshot will contain the list of snapshot ids of the images that comprise the cg's snapshot. Potentially, I assume, we want to allow images to be members of several consistency groups. However it can add complexity to the implementation, see the explanation down the text. The operations that we should be able to perform on consistency groups are: * Create a consistency group * Show a consistency group * List all available consistency groups * Add an image to a consistency group * Create a snapshot of all the members of a consistency group - Created snapshot of a consistency group only has references to snapshots that appear under the corresponding images * List all snapshots of a consistency group * Delete consistency group's snapshot * Modify consistency group (delete, add images to the consistency group) * Enable journaling feature for the whole consistency group. Enabling journaling feature for the whole consistency group will create one journal for all images in the group. All io operations to the images in the group will go to this journal. Every operation will have a tag identifying what image this operation is for. This feature is required for rbd mirroring. In order to make it compatible with cinder consistency groups feature the functionality should cover Cinder Driver API: def create_consistencygroup(self, context, consistencygroup, volumes) def delete_consistencygroup(self, context, consistencygroup) def create_cgsnapshot(self, context, cgsnapshot) def delete_cgsnapshot(self, context, cgsnapshot) https://github.com/openstack/cinder-specs/blob/master/specs/juno/consistency-groups.rst The biggest question regarding the implementation comes from the live snapshot feature. We believe constistent snapshot is not possible without cooperation with rbd client (VM, application), because only application knows if it is in constistent state for snapshotting. In the first iteration we would like to make it simple, so CG snapshot operation just creates snapshots for all images and stores the snapshot list in its metadata. The consistensy should be provided by the user, who should ensure that application is in consistent point (completed all transactions, stopped writting and flushed) during snapshotting. -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
