On 02/21/2011 07:20 PM, Eric Blake wrote: > On 02/21/2011 05:04 PM, KAMEZAWA Hiroyuki wrote: >> On Mon, 21 Feb 2011 08:17:30 -0700 >> Eric Blake <eblake@xxxxxxxxxx> wrote: >> >>> On 02/21/2011 01:23 AM, KAMEZAWA Hiroyuki wrote: >>>> Hi, now, with qemu, virsh attach-disk doesn't work with inactive disks and >>>> we need to edit XML with virsh edit. >>>> IIUC, libvirt and virsh is designed as it is. >>> >>> Actually, libvirt should be patched to learn how to modify xml of >>> inactive disks for qemu (it already can do it for xen, so the API is >>> already present, it's just that no one has wired up that API for qemu). >>> >> >> Before starging this, I thought of that. I did this in python by 3 reasons. >> >> 1. When we asked "Is it a spec that we cannot modify inactive domain ?" to >> a Redhat guy, he answered "it's a spec". >> Do you, maintainers, have some concensus about this ? > > 'virsh attach-disk --persistent' is supposed to be able to modify an > inactive domain. If it doesn't do so for qemu, then that's because no > one has yet implemented it correctly, which means libvirt has a bug that > needs to be patched. For example, see: > > https://bugzilla.redhat.com/show_bug.cgi?id=669549 > > about 'virsh setmem --config' not working for qemu. > Personally I don't think these flags should be implemented anywhere besides the xen driver where they are needed for back compat. Persistently adding devices is something that can be done by an API user in a few dozen lines of code, and it will work for all drivers. If we want virsh to expose this capability, we should just add a generic implementation there. >> >> 2. virsh attach-disk doesn't seem to support misc. options. It doesn't have >> - boot_order >> - shareable >> - serial >> - io >> - error_policy >> etc... > > If there's something that the libvirt API supports, but which virsh does > not support, then that's a bug in virsh. Please let us know about these > usability deficiencies in virsh, since that is the right place to be > patching it for use by all other shell-based tools, rather than > reinventing a new tool by every user. > There are 2 distinct pieces of the libvirt API: API calls and XML building. Both are very large topics in their own right, and I think trying to have a single tool that exposes every nuance of both would be cumbersome. Right now virsh is mostly just an API call wrapper, which is what it should stay IMO. XML building/editing is also closely tied with validation, and higher level operations like creating storage or fixing storage permissions, none of which should live in virsh IMO. Thanks, Cole -- libvir-list mailing list libvir-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/libvir-list