Re: [libvirt] [PATCH 0/3] qemu: libvirt RDMA live migration support
On Fri, Jul 26, 2013 at 12:16:08PM -0600, Eric Blake wrote:
On 07/26/2013 11:47 AM, mrhines(a)linux.vnet.ibm.com wrote:
> From: "Michael R. Hines" <mrhines(a)us.ibm.com>
>
> QEMU has in tree now planned for 1.6 support for RDMA-based live migration.
>
> Changes to libvirt:
> 1. QEMU has a new 'setup' phase in their state machine.
> 2. Expose the 'x-rdma' migration protocol URI.
> 3. Expose the 'x-rdma-pin-all' capability for pre-registration of memory.
The x- prefix means that the migration is still experimental; do we want
to be codifying the use of experimental API into libvirt, or is it time
for a patch to qemu to remove the x- prefix? Back in the 1.5 timeframe,
when RDMA was first proposed, the x- prefix made sense, but now that we
are closer to qemu 1.6, and you are trying to get libvirt to drive it,
that's a declaration of stability.
If it is an experimental API & naming, we definitely do not want to
expose that in the libvirt public API. We need to expose a stable
API for libvirt's client apps. If we can't do that until QEMU itself
is stable, then we'll have to wait.
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 :|