On Thu, 2006-06-29 at 21:49 +0100, Ewan Mellor wrote:
> On Thu, Jun 29, 2006 at 09:32:19PM +0100, Daniel P. Berrange wrote:
> > On Thu, Jun 29, 2006 at 09:11:25PM +0100, Ewan Mellor wrote:
> > > On Wed, Jun 28, 2006 at 09:47:10PM +0100, Daniel P. Berrange wrote:
> > >
> > > > Currently XenD/XenStore doesn't provide a real UUID for Domain-0 - its
> > > > always fixed at 00000000-0000-0000-0000-000000000000. If the new API is
> > > > to allow management of multiple hosts having all their Dom-0's with a
> > > > null UUID is going to complicate things somewhat for users of the API.
> > >
> > > Yes, that's a good point. The reason we did that initially was to ensure
> > > that
> > > your domain 0 UUID didn't change across a reboot (there was a reason we
> > > needed
> > > that, but I can't remember what it was right now).
> > >
> > > I think that we could arrange for domain 0 to get a real but fixed UUID.
> > >
> > > > Is it intended to expose the real UUID of the host being managed. Most
> > > > machines I've encountered have a UUID embedded in the SMBIOS which would
> > > > be the obvious thing to use for Domain-0
> > > >
> > > > $ lshal | grep smbios.system.uuid
> > > > smbios.system.uuid = 'EF861801-45B9-11CB-88E3-AFBFE5370493' (string)
> > >
> > > Well, the host itself needs a UUID too -- wouldn't the SMBIOS UUID be more
> > > appropriate for the host, rather than domain 0?
> > Well if the fully virt guests are going to have their have UUID exposed
> > via SMBIOS, and bare metal OS already uses SMBIOS, then Domain-0 should
> > be consistent, hence SMBIOS too (yes I'm conveniently ignoring paravirt,
> > which has no SMBIOS at this time) Why would we need different UUID for
> > the host, vs Domain-0 OS ? My concern existing userspace tools requiring
> > a UUID for the host/OS currently use SMBIOS, so its desirable to have
> > then 'just work' in virtualized env without changes.
My patches for HVM domU's are progressing again finally. Trying to send
out by tomorrow evening.
> I think that the host and domain 0 are significantly different things, and
> certainly should have different UUIDs. It would be ever so confusing if there
> was an instance of Vm and an instance of Host with the same UUID.
Hrm, this is interesting. Redundant control partitions would present a
real problem if the hardware UUID was used to identify each of them.
It's easy to generate UUID's -- storing a uuid for each control
partition in xenstore or a file would probably work fine, as long as
out-of-band tools can figure out which host maps to each control
partition's uuid. This is problematic without more agent-like code,
unless xend helps with the correlation.
> xen-api mailing list
xen-api mailing list