VM stuck in unshelving

[Ian Kumlien]

VM:s stuck in unshelving has no way forward.

They are stuck in limbo and require database hacks to workaround.

There has to be something a admin can try to restart or reset the unshelve process.

(running on openstack yoga, installed via kolla-ansible)

[Artom Lifshitz (notartom)]

I agree that the reset state API [1] is too limited for cases like you describe, as it can only set the instance to ACTIVE or ERROR.

However, to continue beyond that point we need more information about what happened. Did the instance fail to unshelve and is already in ERROR? Is it stuck in the 'unshelving' task state? In what state is the libvirt domain (I'm assuming the libvirt driver is in use), if one even exists? This mode of failure can be complex, and a simple solution like just resetting state back to SHELVED_OFFLOADED is probably not enough.

I suspect there's a feature request here, but we really need more information about what happened. Triaging this as Incomplete for now.

[1] https://docs.openstack.org/api-ref/compute/#reset-server-state-os-resetstate-action

[Ian Kumlien (pomac)]

it is stuck in unshelving state, so no machine is assigned and there is no indication of it running on any machine... so checking any local state is kinda impossible...

[Launchpad Janitor (janitor)]

[Expired for OpenStack Compute (nova) because there has been no activity for 60 days.]

[Ian Kumlien (pomac)]

And yet another vm stuck in "shelved" that will have to be binned due to broken state...