UEFI NVRAM lost on cold migration or resize

I see the StarlingX fix for this here:

https://github.com/starlingx-staging/stx-nova/commit/71acfeae0d1c59fdc77704527d763bd85a276f9a#diff-f4019782d93a196a0d026479e6aa61b1R8514

I'll bet this would also affect a live migrated instance if it was rebooted.

For a full fix I think you also would need something like https://github.com/starlingx-staging/stx-nova/blob/71acfeae0d1c59fdc77704527d763bd85a276f9a/nova/virt/libvirt/guest.py#L283 in order to keep the nvram around since there are a number of cases where nova is undefining the domain in libvirt but is going to immediately re-create a new domain as the "same" instance and it'd be good to preserve the NVRAM.

Originally we also modified libvirt to store the NVRAM file in the same directory as the other instance files, but I think we're planning on dropping that change since there's going to be no way to override the location of the software TPM file (which means we'll have to deal with that hassle anyway).

Is this related to (or even a duplicate of) bug #1633447 (nova stop/start or reboot --hard resets uefi nvram...) ?

I think it would be strongly related. I'm not sure if it's a dupe, but the solution should be considered with both in mind.

The issue in all cases is that nova implicitly asks libvirt to create the NVRAM file on instance startup but doesn't ensure that it is preserved over all the nova operations.

In this case the issue is that it's not preserved over operations which result in the instance being moved from one node to another. In bug #163344 it's not preserved over instance operations on the same node.

Fix proposed to branch: master
Review: https://review.openstack.org/621646