Cinder did not update '_name_id' and 'provider_location' for drivers if driver implements update_migrated_volume

[Peter Wang]

Currently, Cinder will update '_name_id' and 'provider_location' when drivers don't implement update_migrated_volume
when host-assisted migration has done.

when drivers implemented 'update_migated_volume' but do not return {'_name_id': 'xxxx', 'provider_location': 'xxxxx'},
Cinder will not copy '_name_id' and 'provider_location' to the migrated volume, fields of _name_id and provider_location in migrated volume are left empty.

Above behavior seems break previous host-assisted migration behavior.

since '_name_id' and 'provider_location' are critical for drivers to locate the underlying LUNs in the back-ends, and for keeping back-port compatiblity, we need to copy theses values anyway, and in this way.

here is the code snippets from cinder/volume/manager.py

    def update_migrated_volume(self, ctxt, volume, new_volume,
                               volume_status):
        """Finalize migration process on backend device."""
        model_update = None
        try:
            model_update = self.driver.update_migrated_volume(ctxt,
                                                              volume,
                                                              new_volume,
                                                              volume_status)
        except NotImplementedError:
            # If update_migrated_volume is not implemented for the driver,
            # _name_id and provider_location will be set with the values
            # from new_volume.
            model_update = {'_name_id': new_volume['_name_id'] or
                            new_volume['id'],
                            'provider_location':
                            new_volume['provider_location']}
        if model_update:
            self.db.volume_update(ctxt.elevated(), volume['id'],
                                  model_update)
            # Swap keys that were changed in the source so we keep their values
            # in the temporary volume's DB record.
            model_update_new = {key: volume[key]
                                for key in model_update.iterkeys()}
            self.db.volume_update(ctxt.elevated(), new_volume['id'],
                                  model_update_new)

this behavior is introduced by :

https://review.openstack.org/#/c/180873/

Another problems of above code snippet is
if driver return 'metadata' or 'admin_metadata' to update meta table

            model_update_new = {key: volume[key]
                                for key in model_update.iterkeys()}
will throw exception in attached log since volume does not have 'metadata' or 'admin_metadata'

I am going to provide a fix for this soon.

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/217522
Committed: https://git.openstack.org/cgit/openstack/cinder/commit/?id=20ac9cac7284f009929cfd6e5ce79f34bccffce8
Submitter: Jenkins
Branch: master

commit 20ac9cac7284f009929cfd6e5ce79f34bccffce8
Author: peter_wang <email address hidden>
Date: Fri Aug 21 01:33:44 2015 -0400

    Fix issue of volume after host-assisted migration

    In host-assisted migration, Cinder framework currently doesn't
    update name_id and provider_location when the driver doesn't provide
    name_id or provider_location in update_migrated_volume

    This changes previous Cinder behavior which will update name_id and
    provider_location anyway. Any operation against this volume will not
    be successful; for example, volume deletion cannot delete it's
    underlying LUN.

    This fix will update name_id and provider_location if driver doesn't
    return them.

    Closes-Bug: #1487952
    Change-Id: Ice72fc792f0ea8909c027e6df99cff9bbeef5e48

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/259937

[Peter Wang (peter.wang)]

Please ignore #3, Thanks