Thank you for the answers.

I understood the concerns about having the UUID completely user defined and I also understand Nova has no interest in supporting a customized algorithm to generate UUID. Anyway I may have found a solution that will cover my use case and respect the standard for UUID (RFC 4122 http://www.ietf.org/rfc/rfc4122.txt) .

The generation of the UUID in Nova make use of the function /uuid4()/ from the module /uuid.py/ to have an UUID (pseudo)random, according to version 4 described in RFC 4122. Anyway this is not the only algorithm supported in the standard (and implemented yet in /uuid.py/).

In particular I focused my attention on UUID version 1 and the method /uuid1(node=None, clock_seq=None)/ that allows to pass as parameter a part of the UUID (/node/ is the field containing the last 12 hexadecimal digits of the UUID).

So my idea was to give the chance to the user to set uiid version (1 or 4, with the latter as default) when creating a new instance and in case of version 1 to pass optionally a value for parameter /node/.

Any thoughts?

On 09/30/14 14:07, Andrew Laski wrote:

On 09/30/2014 06:53 AM, Pasquale Porreca wrote:
Going back to my original question, I would like to know:

1) Is it acceptable to have the UUID passed from client side?

In my opinion, no. This opens a door to issues we currently don't need to deal with, and use cases I don't think Nova should support. Another possibility, which I don't like either, would be to pass in some data which could influence the generation of the UUID to satisfy requirements.

But there was a suggestion to look into addressing your use case on the QEMU mailing list, which I think would be a better approach.


2) What is the correct way to do it? I started to implement this feature, simply passing it as metadata with key uuid, but I feel that this feature should have a reserved option rather then use metadata.


On 09/25/14 17:26, Daniel P. Berrange wrote:
On Thu, Sep 25, 2014 at 05:23:22PM +0200, Pasquale Porreca wrote:
This is correct Daniel, except that that it is done by the virtual
firmware/BIOS of the virtual machine and not by the OS (not yet installed at
that time).

This is the reason we thought about UUID: it is yet used by the iPXE client to be included in Bootstrap Protocol messages, it is taken from the <uuid> field in libvirt template and the <uuid> in libvirt is set by OpenStack; the only missing passage is the chance to set the UUID in OpenStack instead to
have it randomly generated.

Having another user defined tag in libvirt won't help for our issue, since it won't be included in Bootstrap Protocol messages, not without changes in the virtual BIOS/firmware (as you stated too) and honestly my team doesn't
have interest in this (neither the competence).

I don't think the configdrive or metadata service would help either: the OS on the instance is not yet installed at that time (the target if the network boot is exactly to install the OS on the instance!), so it won't be able to
mount it.
Ok, yes, if we're considering the DHCP client inside the iPXE BIOS
blob, then I don't see any currently viable options besides UUID.
There's no mechanism for passing any other data into iPXE that I
am aware of, though if there is a desire todo that it could be
raised on the QEMU mailing list for discussion.


Regards,
Daniel



_______________________________________________
OpenStack-dev mailing list
OpenStack-dev@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev

--
Pasquale Porreca

DEK Technologies
Via dei Castelli Romani, 22
00040 Pomezia (Roma)

Mobile +39 3394823805
Skype paskporr

_______________________________________________
OpenStack-dev mailing list
OpenStack-dev@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev

Reply via email to