On Mon, Oct 01, 2018 at 14:58:11 +0200, Marc Hartmayer wrote:
On Mon, Oct 01, 2018 at 12:36 PM +0200, Peter Krempa
<pkrempa(a)redhat.com> wrote:
[...]
>> Yep, that’s why I said in the cover letter “With this patch
series the
>> behavior is still not (completely) fixed, but the performance has been
>> significantly improved.”.
>
> I don't see what should be fixed here, but mostly as I did not read the
> series.
The problem is that we currently don’t take into account that the QEMU
binary can change during a task (e.g. define a domain). Therefore, it’s
possible that two calls of virQEMUCapsCacheLookup return different QEMU
capabilities.
I don't think that is too big of a problem in the light that the binary
can change between the final call to virQEMUCapsCacheLookup which is
used by the code generating the commandline and actual exec of the qemu
binary once the commandline is generated.
Solving that won't be easy at all.
Since virDomainDefValidate is called with the final capabilities
instance when starting the qemu process the possibility that it would
change and the problems it might create are negligible when compared to
actually execing something else.