On Fri, 17 Feb 2012 15:51:33 -0600 Michael Roth <mdr...@linux.vnet.ibm.com> wrote:
> On Fri, Feb 17, 2012 at 03:16:22PM -0200, Luiz Capitulino wrote: > > On Fri, 17 Feb 2012 10:55:41 -0600 > > Anthony Liguori <aligu...@us.ibm.com> wrote: > > > > > On 02/08/2012 02:30 PM, Luiz Capitulino wrote: > > > > This commit adds a QMP API for the guest provided memory statistics > > > > (long disabled by commit 07b0403dfc2b2ac179ae5b48105096cc2d03375a). > > > > > > > > The approach taken by the original commit > > > > (625a5befc2e3200b396594f002218d235e375da5) was to extend the > > > > query-balloon command. It introduced a severe bug though: query-balloon > > > > would hang if the guest didn't respond. > > > > > > > > The approach taken by this commit is asynchronous and thus avoids > > > > any QMP hangs. > > > > > > > > First, a client has to issue the balloon-get-memory-stats command. > > > > That command gets the process started by only sending a request to > > > > the guest, it doesn't block. When the memory stats are made available > > > > by the guest, they are returned to the client as an QMP event. > > > > > > > > Signed-off-by: Luiz Capitulino<lcapitul...@redhat.com> > > > > > > Do we need this to be stable in 1.1? > > > > Well, this is disabled for a long time already and libvirt needs it, so I'd > > say asap, but isn't it possible to implement this with current QOM? > > > > > We can do this pretty nicely through QOM. We can have a polling property > > > in the > > > virtio-balloon driver, that when set, will enable the virtio-balloon > > > device to > > > poll the guest for statistics. > > > > > > > > > We can also have properties for each of the memory statistics and a > > > timestamp > > > for when the last update was. > > > > > > I think this is a friendlier approach for clients, and a cleaner approach > > > from a > > > QEMU perspective. > > > > I agree it's friendlier, but is it a good idea to keep polling the guest for > > something that may never be needed by a mngt app (real question)? > > Probably not, but then again you'd only need like 1-second granularity. I've talked with Anthony by irc about the implementation details of this and it will be possible to enable/disable the polling, so this is not an issue anymore. > Also, I think we can do away with the polling once async QMP is in > place, so we wouldn't be stuck with it necessarilly. This is what this series does :) I don't think it's necessary to wait for async support, we're accepting ad-hoc async mechanisms for other commands and could use one here too if needed. > > We could allow the mngt app to do the polling by adding a > > query-balloon-stats > > command (instead of balloon-get-memory-stats & event). This command could > > return the latest available stats if any (with a timestamp) and query the > > guest for new stats. > > The downside there is you could read some really stale data that way, > to the point where any app that really cared would likely throw out the first > result. Having stale data will be possible with any timer based polling...