[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [libvirt] [PATCH]: Add a flag to let live migration be persistent



On Wed, May 20, 2009 at 11:15:13AM +0100, Daniel P. Berrange wrote:
> On Wed, May 20, 2009 at 12:06:34PM +0200, Daniel Veillard wrote:
> > On Wed, May 20, 2009 at 11:15:14AM +0200, Chris Lalancette wrote:
> > > As pointed out by ??ukasz Mierzwa <l mierzwa gmail com>, it would be nice if
> > > there was an option to automatically make a domain persistent on the destination
> > > during a live migration.  The attached patch adds this simple capability.  Note
> > > that this has to be applied on top of my previous secure migration patch,
> > > otherwise you'll have conflicts.
> > > 
> > > Signed-off-by: Chris Lalancette <clalance redhat com>
> > 
> >   Patch looks fine to me, but
> > 
> > > @@ -5291,6 +5292,32 @@ qemudDomainMigrateFinish2 (virConnectPtr dconn,
> > >       * object, but if no, clean up the empty qemu process.
> > >       */
> > >      if (retcode == 0) {
> > > +        if (flags & VIR_MIGRATE_PERSISTENT) {
> > > +            if (vm->persistent)
> > > +                newVM = 0;
> > > +            vm->persistent = 1;
> > > +
> > > +            if (virDomainSaveConfig(dconn, driver->configDir, vm->def) < 0) {
> > > +                /* Hmpf.  Migration was successful, but making it persistent
> > > +                 * was not.  If we report successful, then when this domain
> > > +                 * shuts down, management tools are in for a surprise.  On the
> > > +                 * other hand, if we report failure, then the management tools
> > > +                 * might try to restart the domain on the source side, even
> > > +                 * though the domain is actually running on the destination.
> > > +                 * Return a NULL dom pointer, and hope that this is a rare
> > > +                 * situation and management tools are smart.
> > > +                 */
> > > +                vm = NULL;
> > > +                goto cleanup;
> > > +            }
> > 
> >   Yup that's a hard decision here, I wonder if some asynchronous event
> > shoul not be generated, at least make sure an error is propagated in
> > some ways.
> 
> I think it is preferable to always return the VM object in this scenario,
> because it is more important for the mgmt tool to know that the migration
> itself has completed successfully and the VM is now running on the dest. 

  yes but an ansynch error would allow the management tools to notice
something happened. If you migrate because you're about to discard an
old server, loosing the domain definition would be nasty.

> I think there is an argument for adding an explicit API
> 
>    virDomainIsPersistent(virDomainPtr)
> 
> to allow an app to determine whether it has a config or not. Even if
> the mgmt tool did not set the 'VIR_MIGRATE_PERSISTENT' flag, they 
> may still need to find out whether it is persistent or not.

  Definitely ! Good point, we need to add this,

Daniel

-- 
Daniel Veillard      | libxml Gnome XML XSLT toolkit  http://xmlsoft.org/
daniel veillard com  | Rpmfind RPM search engine http://rpmfind.net/
http://veillard.com/ | virtualization library  http://libvirt.org/


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]