Re: blueprint: consistency groups

[Jason Dillaman <jdillama@redhat.com>]

On Tue, May 10, 2016 at 12:25 PM, Victor Denisov <vdenisov@mirantis.com> wrote:
> I was writing unit tests for create_cg function when I realized that
> it wouldn't recover if we lost connection between
> cls_client::dir_add_cg and create_cg object. If create_cg doesn't
> happen then we are unable to repeat dir_add_cg.
> However since I implemented create_cg operation after create_image
> operation. create_image operation has the same issue.

Correct - the recovery is to remove the image and re-create if you
have a failure creating the image.  I think that is a valid solution
to the consistency groups as well.

> So, my question is. Should I reimplement create_cg so that it can
> recover from connection loss to the cluster or should I keep it the
> same way create_image is implemented?

There will not be an issue if the connection is lost -- there will
only be an issue if the rbd CLI (or whatever else is using the librbd
API) crashes / is stopped mid-operation.  I am fine with matching the
create_image behavior because there is a recovery path.

> Also, what is more preferable big pull request for all components of
> the feature - CRUD for consistency groups and snapshots, or split it
> into smaller pull requests - first merge CRUD operations then
> implement snapshots?

Smaller patchsets are easier to review, test, and merge (especially as
review comments are addressed).

-- 
Jason