[libvirt-users] Incorrect currentMemory for qemu-1.2+

[Andrey Korolyov]

On Wed, Dec 5, 2012 at 4:44 PM, Andrey Korolyov <andrey at xdel.ru> wrote:
> Hi,
>
> Since this message(https://www.redhat.com/archives/libvirt-users/2012-October/msg00054.html)
> qemu folks have released one minor version in the 1.2 branch and shiny
> new 1.3 some days ago and the problem still persists. At first I has
> blamed libvirt, but I was not correct at all - since I have released
> package ``slices'' for myself including both newer libvirth and qemu,
> I wasn`t able to corner this problem and my report may be counted as
> completely wrong.
>
> Right now, I have the following situation:
>
> - all libvirt versions running above qemu-1.1 always reports right
> balloon and currentMemory sizes
>
> - for qemu 1.2 it is quite more funny: as in message above, I receive
> wrong currentMemory always set to ``magic'' 5398528kB at the 8gb host
> node, but only if ``memory'' parameter has same or higher value, even
> <memory unit='KiB'>5348520</memory> works okay, providing righ balloon
> value for the libvirt. As you can see the ~2700MB offset between
> current and maximum(or entire host) memory present in both
> configurations. 8Gb node is a simple non-numa machine so this is not
> related to numa miscalculation. Previous test at the link above
> processed on the 64GB two-head node with symmetric numa topology, if
> it matters.
>
> And finally,
>
> $ virsh dumpxml vm010-000-000-092 | grep curr
>   <currentMemory unit='KiB'>5398528</currentMemory>
> $ virsh qemu-monitor-command --hmp vm010-000-000-092 info balloon
> balloon: actual=1176
>
> Until virtio_balloon initializes in the guest, libvirt reporting
> _right_ value for some seconds after VM start, so bug (which I unable
> to track by myself) almost for sure presented in parsing of received
> values from guest balloon.

Sorry, forgot to add - in both cases I have tested small vms with size
about 1Gb of RAM. On 8G node raising currentMemory to 2.5G and higher
results to right values, but for 64G machine it was not the same -
even 10Gb value had been reported wrong. At the first glance overall
amount of the host memory also may stepped into this problem, any
suggestions are welcome!