Re: [Qemu-devel] dpdk/vpp and cross-version migration for vhost

[Maxime Coquelin]

On 12/09/2016 03:42 PM, Daniel P. Berrange wrote:
> On Fri, Dec 09, 2016 at 02:35:58PM +0100, Maxime Coquelin wrote:
> > ++Daniel for libvirt
> >
> > On 11/24/2016 07:31 AM, Yuanhan Liu wrote:
> > > > > > > > > > As version here is an opaque string for libvirt and qemu,
> > > > > > > > > > > > > > > > > anything can be used - but I suggest either a list
> > > > > > > > > > > > > > > > > of values defining the interface, e.g.
> > > > > > > > > > > > > > > > > any_layout=on,max_ring=256
> > > > > > > > > > > > > > > > > or a version including the name and vendor of the backend,
> > > > > > > > > > > > > > > > > e.g. "org.dpdk.v4.5.6".
> > > > > > >
> > > > > > > The version scheme may not be ideal here. Assume a QEMU is supposed
> > > > > > > to work with a specific DPDK version, however, user may disable some
> > > > > > > newer features through qemu command line, that it also could work with
> > > > > > > an elder DPDK version. Using the version scheme will not allow us doing
> > > > > > > such migration to an elder DPDK version. The MTU is a lively example
> > > > > > > here? (when MTU feature is provided by QEMU but is actually disabled
> > > > > > > by user, that it could also work with an elder DPDK without MTU support).
> > > > > > >
> > > > > > >         --yliu
> > > > >
> > > > > OK, so does a list of values look better to you then?
> > > Yes, if there are no better way.
> > >
> > > And I think it may be better to not list all those features, literally.
> > > But instead, using the number should be better, say, features=0xdeadbeef.
> > >
> > > Listing the feature names means we have to come to an agreement in all
> > > components involved here (QEMU, libvirt, DPDK, VPP, and maybe more
> > > backends), that we have to use the exact same feature names. Though it
> > > may not be a big deal, it lacks some flexibility.
> > >
> > > A feature bits will not have this issue.
> >
> > I initially thought having key/value pairs would be more flexible, and
> > could allow migrating to another application if compatible (i.e. from
> > OVS to VPP, and vice versa...) without needing synchronization between
> > the applications.
> >
> > But Daniel pointed me out that it would add a lot of complexity on
> > management tool side, as it would need to know how to interpret these
> > key/value pairs. I think his argument is very valid.
> >
> > So maybe the best way would be the version string, letting the
> > application (OVS-DPDK/VPP/...) specify which version it is
> > compatible with.
> > For the downsides, as soon as a new feature is supported in vhost-user
> > application, the new version will not be advertised as compatible with
> > the previous one, even if the user disables the feature in Qemu (as
> > pointed out by Yuanhan).
>
> We need two distinct capabilities in order to make this work properly.
>
> First, libvirt needs to be able to query the list of (one or more)
> supported versions strings for a given host.

Shouldn't be the role of OpenStack/Neutron? IIUC, libvirt knows nothing
about OVS.

> Second, when launching QEMU we need to be able to specify the desired
> version against the NIC backend.
>
> So, consider host A, initially supporting "ovsdpdk-v1". When libvirt
> launches the VM it will specify 'ovsdpgk-v1' as the desired version
> string to use.
>
> Now some time later you add features X, Y & Z to a new release of
> DPDK and install this on host B.  Host B is able to support two
> versions 'ovsdppk-v1' and 'ovsdpdk-v2'.  When libvirt launches
> a VM on host B, it'll pick 'ovsdpgk-v2' by default, since that's
> the newest.   When libvirt migrates a VM from host A, however,
> it will request the old version 'ovsdpdk-v1' in order to ensure
> compatibility.  Similarly when launching a new VM on host B,
> libvirt could choose to use 'ovsdpdk-v1' as the version, in
> order to enable migration to the olver host A, if desired.
>
> This is exactly the way QEMU machine types work, hiding the
> existance of 100's low level settings / default values, that
> a mgmt app would otherwise have to worry about.

I agree on the principle. I need to check what is missing for OVS to
support different versions on different vhost-user ports.

Thanks,
Maxime
> Regards,
> Daniel