On Mon, Feb 15, 2016 at 10:26:17AM +0100, Paolo Bonzini wrote:
>
>
> On 15/02/2016 10:12, Kevin Wolf wrote:
> > Am 14.02.2016 um 17:12 hat Paolo Bonzini geschrieben:
> >> On 14/02/2016 16:52, Paolo Bonzini wrote:
> >>> Reproducer:
> >>>
> >>> x86_64-softmmu/qemu-system-x86_64 \
> >>> -incoming tcp:localhost:12345 -snapshot \
> >>> /vm/virt_test/images/jeos-21-64-base.qcow2
> >>>
> >>>
> >>> Weird as it may seem, this actually makes some sense for testing
> >>> migration with non-shared storage...
> >>
> >> In fact, migration with non-shared storage (migrate -b) is broken in
> >> general, even without -snapshot.
> >
> > Oh cool, does this mean that we can finally remove it? :-)
>
> I wouldn't object, but people are using it. :)
One case I know for sure: Upstream libvirt (consequently, OpenStack
Nova) falls back to 'migrate -b'/("inc:" true) approach when there's no
NBD support on the destination during live block migration:
# From qemuMigrationRun() function in libvirt source
$ less libvirt/src/qemu/qemu_migration.c
[...]
if (migrate_flags & (QEMU_MONITOR_MIGRATE_NON_SHARED_DISK |
QEMU_MONITOR_MIGRATE_NON_SHARED_INC)) {
if (mig->nbd) {
/* This will update migrate_flags on success */
if (qemuMigrationDriveMirror(driver, vm, mig,
spec->dest.host.name,
migrate_speed,
&migrate_flags,
nmigrate_disks,
migrate_disks,
dconn) < 0) {
goto cleanup;
}
} else {
/* Destination doesn't support NBD server.
* Fall back to previous implementation. */
VIR_DEBUG("Destination doesn't support NBD server "
"Falling back to previous implementation.");
}
}
[...]
This fallback also kicks in when people enable TUNNELLED migration (as
NBD does not have support for it in libvirt, due to valid technical
reasons documented else where).
--
/kashyap
