04.05.2016 13:09, Daniel P. Berrange пишет:
On Thu, Apr 28, 2016 at 08:45:11PM +0300, Maxim Nestratov wrote:
> 28.04.2016 20:29, Daniel P. Berrange пишет:
>
>> On Thu, Apr 28, 2016 at 07:57:14PM +0300, Olga Krishtal wrote:
>>> In this proposal description we would like to introduce a separate pool
>>> type:fspool. Fspool provides and manages filesystems.
>>> Below are listed possible examples of fspool, depending on source type:
>>> volume, directory or network.
>> I'm struggling a little to understand what the overal purpose is
>> here. Can you step back a level and describe the overall problem
>> you're facing and what your use cases / goals are for this proposal.
> The main purpose of this is to have a generic API to manage filesystems
> similar to storage API and be able to use these filesystems in domain XMLs
> defining containers.
> So ,it is a kind of abstraction of filesystem sources and backends. Also
> there can be some indirect benefits but they are not the aim.
Ok, I think i see the point now. You'd have a filesystem pool as a object
which manages a bunch of filesystem instances, each one of which can be
used as the backing store for a container filesystem, or for machine virt
filesystem passthrough.
Exactly.
One might think of bolting this onto the directory based storage pool
to let it deal with volumes which are trees instead of flat files, but
I agree that it'd be nicer to have a dedicate set of APIs for filesystem
mgmt.
We thought about this option and came to the same conclusion.
I hope Olga will get back to the mailing list with the first RFC version
of fs pool API shortly.
Thanks for your opinion Daniel.
Maxim
Regards,
Daniel