Live migration failed when the instance is booted with volume and config drive

[Tao Li]

Description
===========

Volume-backed instances fails to live-migrate when config-drive is enabled, the config drive uses iso9660 format. When the instance booted from volume, the libvirt raised an exception: pre-creation of storage targets for incremental storage migration is not supported. In live-migration process, nova considers this instance to live-migrate with block-migrate, but the config drive is copied from source host to dest host in live-migration, and no local disk to migrated by libvirt, so libvirt raised this exception. I think this is a bug. If there is no local disk to migrate by libvirt, we shouldn't use block-migrate.

Steps to reproduce
==================
1. Boot a instance with volume and configdrive.
2. Live-migrate the instance with block-migrate=auto

Expected result
===============
The instance live-migrate successfully.

Actual result
=============
The instance failed to live-migrate, and livrit raised an exception. The message is pre-creation of storage targets for incremental storage migration is not supported.

Environment
===========

Openstack verison: Pike
Livirt version: 3.2.0
Qemu version: 2.9.0

In nova.conf
[libvirt]
images_type = raw

[Matt Riedemann (mriedem)]

Does it work if the config drive format is vfat?

https://docs.openstack.org/nova/latest/configuration/config.html#DEFAULT.config_drive_format

What version of libvirt / qemu do you have?

[Tao Li (eric-litao)]

I haven't tested it when the config drive format is vfat.

The environment information：

Openstack verison: Pike
Livirt version: 3.2.0
Qemu version: 2.9.0

In nova.conf
[libvirt]
images_type = raw

[Tao Li (eric-litao)]

I saw the latest Pike stable have fixed this issue in https://github.com/openstack/nova/blob/3da97201757c81574e3b03108c95cb29e7d7eba3/nova/virt/libvirt/guest.py#L662, so we can close this bug.

[Artom Lifshitz (notartom)]

Closing, as per comment #3.