Open Stack

On Wed, Sep 24, 2014 at 08:26:44AM +1000, Michael Still wrote:
> On Tue, Sep 23, 2014 at 8:58 PM, Daniel P. Berrange <berrange at redhat.com> wrote:
> > On Tue, Sep 23, 2014 at 02:27:52PM +0400, Roman Bogorodskiy wrote:
> >>   Michael Still wrote:
> >>
> >> > Hi.
> >> >
> >> > I know we've been talking about deprecating nova.virt.disk.vfs.localfs
> >> > for a long time, in favour of wanting people to use libguestfs
> >> > instead. However, I can't immediately find any written documentation
> >> > for when we said we'd do that thing.
> >> >
> >> > Additionally, this came to my attention because Ubuntu 14.04 is
> >> > apparently shipping a libguestfs old enough to cause us to emit the
> >> > "falling back to localfs" warning, so I think we need Ubuntu to catch
> >> > up before we can do this thing.
> >> >
> >> > So -- how about we remove localfs early in Kilo to give Canonical a
> >> > release to update libguestfs?
> >> >
> >> > Thoughts appreciated,
> >> > Michael
> >>
> >> If at some point we'd start moving into getting FreeBSD supported as a
> >> host OS for OpenStack, then it would make sense to keep localfs for that
> >> configuration.
> >>
> >> libguestfs doesn't work on FreeBSD yet. On the other hand, localfs
> >> code in Nova doesn't look like it'd be hard to port.
> >
> > Yep, that's a good point and in fact applies to Linux too when considering
> > the non-KVM/QEMU drivers libvirt supports. eg if your host does not have
> > virtualization and you're using LXC for container virt, then we need to
> > have localfs still be present. Likewise if running Xen.
> >
> > So we definitely cannot delete or even deprecate it unconditionally. We
> > simply want to make sure localfs isn't used when Nova is configured to
> > run QEMU/KVM via libvirt.
> >
> > So if we take the config option approach I suggested, then we'd set a
> > default value for the vfs_impl parameter according to which libvirt
> > driver you have enabled.
> 
> I'm glad we've had this thread, because I hadn't thought of the
> FreeBSD case at all. In that case I wonder if we want to water down
> the warning we currently log in this case:
> 
>                 LOG.warn(_LW("Unable to import guestfs"
>                              "falling back to VFSLocalFS"))
> 
> If feel like it should be an info if we know some platforms will
> always have this occur. I know this is a minor thing, but this came to
> my attention because at lease one operator was concerned by seeing
> that warning in their logs.

If we take my suggested approach of using a fixed impl based on libvirt
driver type, then we wouldn't have fallback & so not see this warning.
Even when we do have fallback, we should only warn if libguestfs is
installed, but not working.


Regards,
Daniel
-- 
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|