On Thu, Jul 04, 2024 at 08:24:30AM +0200, Michal Prívozník wrote: > The fact that domain XML is not stored on a disk doesn't necessarily > mean it can't have inactive XML. Those are two different things. I spent a bit of time playing, and the behaviour was more implemented than I thought for the non-LIVE case; I think I was missing how the vmdefs are retrieved based upon the flags. > > For our application we actually do want to pass the LIVE flag - we never want to > > modify any persistent configuration. We'd prefer not to have to carry a patch on > > top for this. > > Ah, I thought you're interested only in _CONFIG since that's what your > original patch implemented. I took a look at my original patch with fresh eyes, and see it was a bit confused in fact. It does implement LIVE (vm->def) (and I tested that), but picked the wrong function name from the qemu driver sources: 10427 if ((ret = testDomainUpdateDeviceConfig(vm->def, dev, ^^^^^^ ^^^^^^^ 10428 parse_flags, 10429 driver->xmlopt)) < 0) Obviously that makes no sense! > But implementing live update should be > pretty trivial. It's a test driver so "hypervisor" will allow just any > change. Do you want to post a patch for that or should I? Now I have a better idea of what was wrong, let me have another pass. thanks john