[Engine-devel] VmDynamic kvm_enable

Dan Kenigsberg danken at redhat.com
Wed Mar 13 09:22:58 UTC 2013


On Wed, Mar 13, 2013 at 12:58:44AM +0200, Itamar Heim wrote:
> On 03/12/2013 05:07 PM, Dan Kenigsberg wrote:
> >On Tue, Mar 12, 2013 at 10:37:41AM -0400, Laszlo Hornyak wrote:
> >>Hi,
> >>
> >>I came across a VmDynamic property 'kvm_enable'. It sounds strange for me, because ovirt is very colsely integrated with kvm. So a short dig into this flag...
> >>There is a similar thing for the VDS, it is set to true by vdsm is the host CPU has a VT flag. It is actually used to check if the host is OK to run VMS.
> >>
> >>But the one for Vm it looks like a distributed logical loop in vdsm: it is set when constructing a VM object (vdsm/vm.py:~343) from the data sent by the client (engine) and then reported back in vm stats, so it is just a roundtrip between vdsm and it's client.
> >>In the engine side, it is just keeps sending it between the frontend DB and vdsm, never part of a decision.
> >>
> >>Is this still needed here? Can I remove?
> >>VDSM guys?
> >
> >I have a vague memeory that once upon at time, qemu occasionally failed
> >to enable kvm support - even though it was asked to. I then silently
> >switched to emulated mode, which was grindingly slow. Engine wanted to
> >know about such occasions.
> >
> >I believe that a better technical approach would have been to kill the
> >violating process, and not let it run at all (unless qemu emulation was
> >strictly requested by management).
> >
> >Anyway, as you have noted, this has rotten away throuh the years, and
> >unless older Engine versions are expecting this value in any way, I am
> >all for dropping it.
> 
> I think this was about something else - the kvmEnable flag was used
> to launch installations of guests not correctly supported by kvm
> that required real mode or something like that.

You are now speaking about kvmEnable sent by Engine to Vdsm. Laszlo was
speaking about a supposedly-dynamically-changing kvmEnable that is
reported back, in getVmStats.

> hopefully not relevant any more, but need to validate won't break
> older engines indeed.




More information about the Engine-devel mailing list