libvirt xen hvm guests can't hot plug (attach detach) default block devices

[Nikita Gerasimov]

Currently default disk bus for libvirt+Xen HVM guests is "ide", which not allow hot reconfiguration. So images without special metadata property specifying bus can't be attached or detached.

https://github.com/openstack/nova/blob/ad1c7ac2b102112280a25927d731edb168f80998/nova/virt/libvirt/blockinfo.py#L250

At the same time QEMU/KVM guests have "virtio" as default disk bus for disk which is QEMU analog of "xen" bus for XEN.

Fix looks trivial and require only delete XEN HVM specific part to return "xen" as default part in all cases. User will be able to use other buses by metadata and only when it's really required.

[Anthony PERARD (anthony-perard)]

I don't think it is that simple to change the default. I guest still require an "ide" disk in order to boot. The default BIOS on Xen will not be able to boot a guest that only have "xen" disks. So we may need to do something else in order to hot-plugs extra disks.

[Nikita Gerasimov (nikita-gerasimov)]

> I guest still require an "ide" disk in order to boot.
No they do not. Default OS server creation scenario is boot operator prepared images which contain necessary drivers. There is only one case when you really need "ide" bus. When you creating a new template image, and only until you confiure/setup Xen drivers. So "ide" usage is one boot per image. Even less because most of current Linux and *BSD have it by default, even cirros works fine as "xen" bus only guest. In other hand this a rare need of "ide" covered by metadata property. In current state user have to set it manually for all images by default. That looks weird.

> The default BIOS on Xen will not be able to boot a guest that only have "xen" disks.
What Xen version are we talking about? Mine does it fine. Could you please provide more info about your setup?

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/457320

[OpenStack Infra (hudson-openstack)]

Change abandoned by Matt Riedemann (<email address hidden>) on branch: master
Review: https://review.openstack.org/457320
Reason: Based on responses from Bob and Anthony, and those were several months ago now, I'm going to abandon this since it's stuck.