Comment 5 for bug 1627694

@mriedem Thank you for your response.
We don't have to detach the volume from the server, or deleting anything from the DB. The problem is simply that the volumes are stayed connected to the node, because initialize_connection was called, but when spawning failed, and the VM is not on the node, the volumes stay connected to the node.
If terminate_connection is called on rollback, the user can safely retry the unshelving, and the volumes won't be connected to more than 1 node. Nothing will be changed in the DB regarding the VM's attached volumes or networks.