Open Stack

I came across this bug during triage today:

https://bugs.launchpad.net/nova/+bug/1732428

It essentially says that unshelving an instance and then resizing that 
instance later, depending on the type of image backend, can fail.

It's pointed out that when we complete the unshelve procedure, we set 
the instance.image_ref back to the original image_ref used to create the 
instance, rather than leave it pointing at the shelved instance snapshot 
image id.

I thought, "well that's crazy, the instance isn't backed by the original 
image anymore, it's backed by the snapshot image, so instance.image_ref 
should point at the snapshot image id now." But lo, in true 
shelve-tastic form, it turns out that would cause more bugs.

Because after we successfully spawn the guest during unshelve, we delete 
the snapshot image:

https://github.com/openstack/nova/blob/b6a245f0425a07be3871a976952646d2bdd44533/nova/compute/manager.py#L4797

So at this point, you've unshelved your instance, but the 
instance.image_ref is pointing at image A but was really created from 
image B.

Does anyone have ANY idea why we do this? Even if we delete the snapshot 
image, why would we change the image_ref back to the original image?

I think we can assert as follows:

1. After you've unshelved an instance, it's image_ref (unless 
volume-backed, because that's different crazy, not discussed here) 
should point at the image used to create the guest. Agree?

2. If we're going to point the instance at an image_ref, we shouldn't 
delete that image. I don't have a good reason why besides deleting 
things which nova has a reference to in an active instance generally 
causes problems (volumes, ports, etc).

Am I missing something that everyone else knew about way back in 2013?

-- 

Thanks,

Matt