migration 147 wrong, would loose AZ data

[Sean Dague]

Migration 147 turns out not to work when more than one availability zone is specified in services, as the where clause was incorrect in looking for existing AZ definitions.

[Davanum Srinivas (DIMS) (dims-v)]

For the record, the review is https://review.openstack.org/#/c/20664/

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/20664
Committed: http://github.com/openstack/nova/commit/3c64e6ca4cfe4e61348ef1438787fb61d04d5ec7
Submitter: Jenkins
Branch: master

commit 3c64e6ca4cfe4e61348ef1438787fb61d04d5ec7
Author: Sean Dague <email address hidden>
Date: Mon Jan 28 17:51:10 2013 -0500

    refactored data upgrade tests in test_migrations

    The one off testing in test_migrations was starting to get obtuse
    enough that it was possible for bugs to slip through (which they
    did already). Create a more general framework for testing data
    transitions at any db migration.

    Hook the walk_versions code so as we are about to upgrade to a
    migration point, attempt to run a _prerun_### function, then follow
    it up with a _check_### function afterwards. Create some utility
    functions to make doing things in these _prerun and _check functions
    pretty concise.

    This makes these checks db independent, and has it so they'll run
    however we are running the walk_versions.

    In doing so, this uncovered a bug in the 147 migration, which is
    fixed inline. There is no retroactive fix, as 147 was destructive to
    data, so would have nuked anyone that hit it already.

    Updated to remove safe_getattr, thanks to jerdfelt for pointer on
    doing that right.

    Partially implements blueprint migration-testing-with-data

    Fixes bug #1108448

    Change-Id: I8bd92e3688109bfbdf1da21a175055b7263abf45