[virt-tools-list] [virt-manager PATCH] virt-install: refresh storage pool after disk is created

[Pavel Hrdina]

On Tue, Nov 10, 2015 at 04:51:25PM -0500, Cole Robinson wrote:
> On 11/10/2015 08:19 AM, Pavel Hrdina wrote:
> > This fixes an issue where you need to manually refresh storage pool or
> > restart virt-manager in order to be able delete guest that is currently
> > installed.  This affect remote connection and non-root users.
> > 
> > Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1251400
> > 
> > Signed-off-by: Pavel Hrdina <phrdina at redhat.com>
> > ---
> >  virtinst/storage.py | 2 ++
> >  1 file changed, 2 insertions(+)
> > 
> > diff --git a/virtinst/storage.py b/virtinst/storage.py
> > index e50b78f..21d21a6 100644
> > --- a/virtinst/storage.py
> > +++ b/virtinst/storage.py
> > @@ -797,6 +797,8 @@ class StorageVolume(_StorageObject):
> >                  logging.debug("Using vol create flags=%s", createflags)
> >                  vol = self.pool.createXML(xml, createflags)
> >  
> > +            self.pool.refresh()
> > +
> >              self._install_finished = True
> >              t.join()
> >              meter.end(self.capacity)
> > 
> 
> I don't think this patch actually fixes the reported issue... it doesn't do
> anything for me at least.
> 
> The root problem here is that libvirt doesn't support event APIs for storage
> operations, and virt-manager no longer relentlessly polls the storage list
> since it sucks for high latency connections. The disk image is created behind
> virt-managers back, and nothing in virt-manager is catching that it's updated.
> 
> Particularly the pool.refresh() call here isn't going to signal virt-manager
> in any way to update its internal pool/vol cache, so it won't be reflected in
> the app.

Ouch, you are right, I've probably had that disk in virt-manager's cache and
that's why I thought, that was a correct fix.

> 
> We already have some hacks in place to deal with outdated cache in
> virt-manager, like we will request a pool/vol cache update when launching the
> new VM wizard. But in this particular case I don't think it's important enough
> to try and fix, we should just defer it until libvirt has storage event APIs.

Yes, we need to wait for those events and it's obvious that virt-install cannot
affect virt-manager's cache.

Thanks for review,

Pavel

> 
> Thanks,
> Cole