On Tue, 18 Dec 2012 14:34:16 -0700 Eric Blake <ebl...@redhat.com> wrote:
> On 12/17/2012 03:13 AM, Dietmar Maurer wrote: > >> Next commit will re-enable balloon stats with a different interface, but > >> this > >> old code conflicts with it. Let's drop it. > > > > I don't really see any conflicts here? > > > >> It's important to note that the QMP and HMP interfaces are also dropped by > >> this commit. That shouldn't be a problem though, because: > >> > >> 1. All QMP fields are optional > >> 2. This has never been really used > > Libvirt has been using it when available (although reluctantly, as it > risks hanging on an uncooperative guest); This has always been disabled and qemu never returns the stats info. I believe libvirt's code is rotting just like qemu's is. > and while libvirt can be > patched to call 6 QOM commands in a row to query six different QOM > stats, I still think it would be nicer to add a command that provides > all the stats at once. In particular, when calling 6 commands in > series, you no longer have an atomic picture of the guest (the polling > interval could hit between two QOM queries, resulting in a combined set > of statistics that has no counterpart to the transition of states that > the guest actually went through). On the other hand, since the stats > are already polling-based, and since it requires cooperation from the > guest, not having a guarantee of an atomic set of stats is not really > much of a loss. Something I have been wondering if whether it's possible to have only one property (say balloon-statistics) and return all properties in a dict. QOM properties return a visitor, so maybe that's possible. I'll check that.