Hi!

On Tue, Mar 5, 2013 at 6:27 PM, Guido Trotter <[email protected]> wrote:

> On Tue, Mar 5, 2013 at 9:14 AM, Guido Trotter <[email protected]>
> wrote:
> > On Tue, Mar 5, 2013 at 5:18 AM, Helga Velroyen <[email protected]>
> wrote:
> >> I fixed your comments and also added the keys back to the RPC call.
> From my
> >> notes of two weeks ago (I should have written that right away), I get
> that
> >> this was what Guido had in mind: the enabled storage methods as list on
> >> cluster level, but the (method,key) tuple in the call.
> >>
> >
> > Indeed, thanks&LGTM to this: that means that the storage pools
> > implementation can use the RPC, while we don't need to implment those
> > "immediately" while starting to have the first benefits.
> >
>
> Quick sync with iustin:
>
> - It's ok to change the RPC as said
> - For gnt-node list we can follow your proposal, and when we insert
> storage pool, we'll call by default the drbd storage pool "drbd", of
> course it will be possible to rename it. gnt-node list will by default
> report on the default storage pool, and of course you'll be able to
> ask about any of them (and there will be a new query to ask "what
> storage pools are there, and of what type).
> - For the iallocator interface, changes should also take care of being
> forward-compatible, which should also be fine as we'll be able to add
> the storage pools later as explicit objects and report space for all
> available ones too (for now we'll report space for all available
> storage types, in the future for all available storage pools).
>


Good points, thank you. Although the design doc technically does not cover
the specifics of the storage pools, I added comments about the
forward-compatibility and relationship to storage pools, so that the
context of the decisions is more clear.

Cheers,
Helga

Reply via email to