Resizing an instance with a volume attached fails to reattach volume to instance

Fix proposed to branch: master
Review: https://review.openstack.org/13071

This is a pretty serious bug, but the fix is fairly scary - e.g. starting the VM as paused and then resuming it is the kind of thing that is likely to break in some weird way in some obscure situations

This looks like it was always issue and isn't a regression since Essex. Since rc1 is about to be cut within hours, it makes sense to defer this fix until Grizzly. It can always be backported to stable/folsom later

If I can eliminate the need to start the VM paused, would you reconsider merging this in for Folsom?

The idea would be to to look into how generate the volume references in the XML prior to starting the VM, rather than calling attach afterwards (not sure how to do this yet).

I've reworked the patch to eliminate pausing VMs paused. Hopefully this mitigates the concerns around potentially obscure situations.

More detail about how/what I changed is in the review comments.

Failing a proper fix in folsum, can we at least ensure that the resize command detects the volume's and refuses to start the resize?

Reviewed: https://review.openstack.org/13071
Committed: http://github.com/openstack/nova/commit/b01ec8012a2a2568feb5fa05a56909a6bdf14ef8
Submitter: Jenkins
Branch: master

commit b01ec8012a2a2568feb5fa05a56909a6bdf14ef8
Author: Rafi Khardalian <email address hidden>
Date: Sun Sep 16 16:28:58 2012 +0000

    Re-attach volumes after instance resize

    Fixes bug 1051600.

    Update the compute manager to pass block_device_info to three
    additional driver entry points and update all virtualization drivers
    to accept it as an optional argument.

    Within libvirt, migrate_disk_and_power_off() will now iterate and disconnect
    any existing connections. finish_migration() simply updates its use
    of to_xml(), passing block_device_info, which returns a libvirt XML
    fully populated with volumes. Finally, finish_revert_migration() no
    longer uses the on-disk XML file, instead generating it via to_xml; as
    the former lacks volume configuration.

    Change-Id: I7dccd03ef9cc7d2848f07764d4def3787e41e202