[libvirt] [PATCH 4/4] qemu_capabilities: Temporarily disable dbus-vmstate capability

Marc-André Lureau marcandre.lureau at redhat.com
Mon Sep 9 11:57:03 UTC 2019


Hi

On Mon, Sep 9, 2019 at 2:49 PM Ján Tomko <jtomko at redhat.com> wrote:
>
> On Mon, Sep 09, 2019 at 11:23:59AM +0200, Michal Privoznik wrote:
> >On 9/6/19 5:55 PM, Ján Tomko wrote:
> >>[cc-ing Marc-André]
> >>
> >>On Fri, Sep 06, 2019 at 04:25:19PM +0200, Michal Privoznik wrote:
> >>>The qemu side is not merged in yet, so there is a chance that the
> >>>interface will change. Don't detect the capability just yet then.
> >>>
> >>>Signed-off-by: Michal Privoznik <mprivozn at redhat.com>
> >>>---
> >>>src/qemu/qemu_capabilities.c | 1 -
> >>>1 file changed, 1 deletion(-)
> >>>
> >>>diff --git a/src/qemu/qemu_capabilities.c b/src/qemu/qemu_capabilities.c
> >>>index 136999ad0d..7be026ae12 100644
> >>>--- a/src/qemu/qemu_capabilities.c
> >>>+++ b/src/qemu/qemu_capabilities.c
> >>>@@ -1129,7 +1129,6 @@ struct virQEMUCapsStringFlags
> >>>virQEMUCapsObjectTypes[] = {
> >>>    { "virtio-serial-pci-non-transitional",
> >>>QEMU_CAPS_VIRTIO_PCI_TRANSITIONAL },
> >>>    { "max-x86_64-cpu", QEMU_CAPS_X86_MAX_CPU },
> >>>    { "bochs-display", QEMU_CAPS_DEVICE_BOCHS_DISPLAY },
> >>>-    { "dbus-vmstate", QEMU_CAPS_DBUS_VMSTATE },
> >>>};
> >>>
> >>>static struct virQEMUCapsStringFlags
> >>>virQEMUCapsDevicePropsVirtioBalloon[] = {
> >>
> >>If
> >>1) dbus-vmstate is required for migration of a domain using a
> >>slirp-helper and
> >>2) we enable this automatically for existing XML configs with
> >>  <interface type='user'/> if
> >>  a) QEMU has QEMU_CAPS_NET_SOCKET_DGRAM
> >>  b) the slirp-helper process is installed on the system
> >>then we've effectively made their machines unmigratable
> >>
> >>So if we want to enable it automatically, we should add:
> >>c) QEMU has QEMU_CAPS_DBUS_VMSTATE
> >>regardless of whether dbus-vmstate gets merged into QEMU until libvirt's
> >>next release
> >
> >So are you saying that we should just temporarily revert
> >QEMU_CAPS_NET_SOCKET_DGRAM detection instead and leave
> >QEMU_CAPS_DBUS_VMSTATE be?
>
> Give that there was a QEMU release which has one but not the other,
> I'd say permanently revert QEMU_CAPS_NET_SOCKET_DGRAM - even if
> dbus-vmstate gets merged later, there has been a QEMU release for which
> we assume one capability and not the other - so libvirt would be
> starting unmigratable machines.
>
> As for QEMU_CAPS_DBUS_VMSTATE, we can either revert it temporarily now
> or before the libvirt freeze if dbus-vmstate does not get merged by
> then.

I agree that as long as qemu doesn't have dbus-vmstate, it is better
to revert dbus-vmstate support, and disable net-socket-dgram.

>
> Jano
>
> >
> >Michal
> >
> >--
> >libvir-list mailing list
> >libvir-list at redhat.com
> >https://www.redhat.com/mailman/listinfo/libvir-list




More information about the libvir-list mailing list