On Thu, Aug 27, 2015 at 01:27:40PM -0600, Eric Blake wrote: > On 08/27/2015 01:08 PM, Jeff Cody wrote: > > On Thu, Aug 27, 2015 at 05:02:17PM +0100, Daniel P. Berrange wrote: > >> > >> Just to be clear - libvirt will *never* use an auto-generated device > >> IDs feature. It is way more complicated to let QEMU assign device IDs > >> and then auto-detect them from some 'info device-list' output, than > >> to just specify IDs upfront at device/object creation time which > >> it already does[1]. There is simply no benefit to auto-generating device > >> IDs for a mgmt app like libvirt, and plenty of downside. Auto-generated > >> IDs will only be of interest to humans talking to the monitor directly > >> without a mgmt app involved. > >> > >> Regards, > >> Daniel > >> > >> [1] we don't provide IDs for qcow2 image backing file chain, but that's > >> part of a bigger story that's being dealt with in this area. > > > > This is the part that interests me the most :) > > > > Do you think in dealing with image backing file chains, libvirt would > > ever make use of QEMU auto-generated node-names (either in the > > current feature set, or with future features)? I'm not sure if your > > above statement is specific to device ID, or extends to node-names as > > well. > > I have a patch series that I posted for libvirt that used a hack to try > and take advantage of auto-generated names (basically, you can't use > allocation watermark events on qcow2-over-block-devices without proper > node names). Had we turned on auto node names in 2.4 (the release that > added the event), then it might be in libvirt now. But now that we have > to support 2.4 without auto node names, it's in libvirt's long-term > interest to avoid technical debt and directly supply node names for all > BDS, rather than relying on node names, and rather than cheating by > waiting for qemu 2.5. > > So my current task in libvirt is to try and fix things to supply node > names everywhere, then rewrite my allocation watermark event series atop > that change, at which point relying on auto names will not be necessary > for libvirt. >
OK, thanks. That certainly makes it less urgent, at least on my end. Seems like the only time we'd need it for libvirt is if in the future we created BDSs as a byproduct of an operation, and wanted to have a node-name so that we could present it to libvirt afterwards. Thanks, Jeff
