于 2010年12月08日 23:42, Eric Blake 写道:
On 12/06/2010 05:35 AM, Osier Yang wrote:
As qemu driver doesn't allow to make changes on persistent
domain configuration via "attach/detach/update device",
However, this feature has been requested [1] - how easy will it be to
add that support after this patch is applied?
[1] https://bugzilla.redhat.com/show_bug.cgi?id=658713
yeah, it will affects patch for that feature, but not much
I think, just need to skip the block like:
if (!persistent) {
if (!vm->updated)
vm->updated = 1;
......
}
and all the changes made on the running domain configuration
should not be persistent across next boot (without the need
of restarting libvirtd), so:
1) Set the running domain def as transient, and restore
the domain configuration to original configuration when
shutdown.
2) Set the running domain def as updated, and reset it as
not updated when shutdown.
Also for "live VCPU set", it doesn't change the persistent
domain configuration, so, we also set the running domain
def as updated and transient, and restore the original def
when shutdown.
@@ -6353,6 +6360,17 @@ qemudDomainSetVcpusFlags(virDomainPtr dom, unsigned int
nvcpus,
if (flags& VIR_DOMAIN_VCPU_CONFIG)
ret = virDomainSaveConfig(driver->configDir, persistentDef);
+ if (flags& VIR_DOMAIN_VCPU_LIVE) {
+ /* Set running domain def as updated */
+ if (ret == 0) {
+ vm->updated = 1;
+
+ if (virDomainObjSetDefTransient(driver->caps, vm)< 0)
+ VIR_ERROR("Unable to set domain %s's running config as
transient",
+ vm->def->name);
+ }
+ }
In all of these cases, you issue VIR_ERROR but don't change the return
status. Is that intentional, or should we make the overall API fail if
we can't create the transient counterpart?
It was intentional, as probly the domain configuration was updated
successfully by previous codes, but yeah, it's not much reasonable here.
And depending on the answer to that question, should we be attempting
the virDomainObjSetDefTransient sooner, so as to guarantee that we have
the transient counterpart prior to making any attempt to change the
running domain, so that we don't have to worry about rolling back
changes as part of returning failure?
Ah, yeah, good idea, using virDomainObjSetDefTransient sooner will be
avoid considering the upper question, as the running domain
configuration should be transient properly, we even don't need to
worry about whether the coming changes on the config be success or not.
Will update, thanks, Eric
- Osier
--
libvir-list mailing list
libvir-list@redhat.com
https://www.redhat.com/mailman/listinfo/libvir-list
