xenapi: stop races when creating a cached image

[John Garbutt]

Currently, when creating an instanced from a cached image there is a race where we can end up fetching the wrong base image, leading to the creation of a bushier VHD chain that should be possible.

We should us a mutex to stop some of these issues.

[John Garbutt (johngarbutt)]

https://review.openstack.org/#/c/46706/

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/46706
Committed: http://github.com/openstack/nova/commit/ce8e95c7a3fe9d94d6b03ee70c147e5e92aa3b5b
Submitter: Jenkins
Branch: master

commit ce8e95c7a3fe9d94d6b03ee70c147e5e92aa3b5b
Author: John Garbutt <email address hidden>
Date: Fri Sep 13 17:04:06 2013 +0100

    xenapi: stop using get_all_vdis_in_sr in spawn

    Currently when trying to find a cached image, the very expensive call to
    get_all_vdis_in_sr is used, where we could instead fetch the VDI
    directly using a more targeted query.

    Now we are fetching the VDI by name_label we must ensure to clear the
    name_label on newly created VDIs to ensure they do not get picked up by
    later calls to _find_cached_image().

    Fixes bug 1221292

    The above code that checks for _find_cached_images has race conditions
    where its possible to end up with two VDIs returned. To stop this
    happening the code to create the cached images is now synchronized on
    the image being fetched.

    Fixes bug 1226073

    Change-Id: I534fb8f42b00b5d39dc17dd5fee297144b5f379a