[libvirt] RFC: VIR_DOMAIN_SNAPSHOT_LIST_NO_METADATA is not functional?

[Han Han]

On Mon, Nov 5, 2018 at 9:38 PM Peter Krempa <pkrempa at redhat.com> wrote:

> On Fri, Nov 02, 2018 at 15:52:54 +0800, Han Han wrote:
> > Hello,
> > I found snapshot APIs(like virDomainSnapshotNum) invoked with
> > VIR_DOMAIN_SNAPSHOT_LIST_NO_METADATA will return 0 even there are
> internal
> > no-metadata snapshots in the domain.
>
> We've never implemented the support for loading internal snapshots
> without libvirt metadata, thus none will be listed.
>
> >
> > Then I find the reason is in virDomainSnapshotObjListGetNames():
> >    937     /* If this common code is being used, we assume that all
> > snapshots
> >    938     ┆* have metadata, and thus can handle METADATA up front as
> > an
> >    939     ┆* all-or-none filter.  XXX This might not always be true, if
> > we
> >    940     ┆* add the ability to track qcow2 internal snapshots without
> > the
> >    941     ┆* use of metadata.
> > */
> >    942     if ((data.flags & VIR_DOMAIN_SNAPSHOT_FILTERS_METADATA)
> > ==
> >    943     ┆
> > VIR_DOMAIN_SNAPSHOT_LIST_NO_METADATA)
> >    944     ┆   return 0;
> >
> > So currently, with VIR_DOMAIN_SNAPSHOT_LIST_NO_METADATA, no snapshot will
> > be returned.
> >
> > I checked the commit. It is really old(6478ec1673a    Eric Blake
> > 2012-08-13 18:09:12 -0600) . I guess it was initially designed for the
> > function to list internal snapshots without metadata.
> >
> > However, now internal snapshot seems lower prioritized than external
> > snapshot.
> > Do we have plan to design this function? Since the APIs invoked with this
> > flag don't return the **right** result, if the function for
> > VIR_DOMAIN_SNAPSHOT_LIST_NO_METADATA will not be designed, how will we
> deal
> > with the flag? Remove it from code OR update the docs to mark it
> > unsupported?
>
> We cannot remove that flag. It will return no results until somebody
> actually implements the support for snapshots with no libvirt metadata.
>
Since we cannot remove it, it's better to note it in the virsh manual and
code comments
in case it makes users confused about the none result.

> Also note that for other hypervisor drivers there certainly is more
> possibility for modelling that as well so removing the flag does not
> make sense.
>


-- 
Best regards,
-----------------------------------
Han Han
Quality Engineer
Redhat.

Email: hhan at redhat.com
Phone: +861065339333
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/libvir-list/attachments/20181106/65db67db/attachment-0001.htm>