It may be worth taking into account that we had a work item to write a FFU hook for this, but it was never done. http://lists.openstack.org/pipermail/openstack-dev/2018-October/136110.html http://lists.openstack.org/pipermail/openstack-discuss/2019-March/004204.htm... efried_stirs_pot_from_afar On 3/31/20 7:35 AM, Sylvain Bauza wrote:
Hola,
In Stein, I provided a new feature [1] that was checking inventories from multiple children resource providers (RP). Since in Rocky, those inventories were only on the root RP, I had to provide both : #1 : an upgrade support for RPs that were not having a parent RP (that was the case of Stein compute nodes) [2] #2 : a reshape method in the libvirt driver for moving allocations from a root RP to children RPs [3]
Now, we're in the Ussuri timeframe and as I was usually doing, I removed the upgrade support (in #1) since we're 100% sure that all compute nodes are either Train or Ussuri [4]
This is cool, but now the reshape code I wrote for #2 no longer works (which is normal, since the 'non-Stein' support was removed) [5]
Soooooo. Can I also then remove the reshape support ? That would mean that upgrading from Rocky to Train would work, but Rocky to Ussuri wouldn't. I'm personally OK with this, but I guess other people could have concerns with.
Do we have any consensus about how longer we would support a reshape ? I think honestly that a 3-releases time window is enough but I can hear some disagreements about this strawman proposal. If so, please voice.
-Sylvain
[1] https://blueprints.launchpad.net/nova/+spec/vgpu-stein [2] https://review.opendev.org/#/c/636591/ [3] https://review.opendev.org/#/c/599208/ [4] https://review.opendev.org/#/c/715489/ [5] https://541e2403478ac154d5eb-056bfb946e355d1a1a86dc411a70c5ec.ssl.cf2.rackcd...