Poll rescued instances fails with key error

[John Garbutt]

After an instance has been in the rescue state for some time, a periodic task triggers to unrescue the instances:
_poll_rescued_instances

File nova/notifications.py info_from_instance
  instance_type = flavors.extract_flavor(instance_ref)
File "nova/compute/flavors.py" in extract_flavor
  instance_type[key] = type_fn(sys_meta[type_key])
KeyError: 'instance_type_memory_mb'

This then continues to happen on every run of the periodic task, and starts to fill up the DB with instance faults.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/71504

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/71504
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=ea7242a293abe59dd91add5e44e58a2eb6cd99fc
Submitter: Jenkins
Branch: master

commit ea7242a293abe59dd91add5e44e58a2eb6cd99fc
Author: John Garbutt <email address hidden>
Date: Thu Feb 6 11:09:33 2014 +0000

    Fix auto instance unrescue after poll period

    We were seeing massive amounts on instance faults from failed automatic
    unrescue. It was coming from the instance not having system metadata
    populated.

    While the route cause is a combination of using non-object instances
    that were not populated with system metadata. With an object, the system
    metadata would lazyload as required, but clearly not with a pure dict.

    This fix ensures the non-object instance has system_metadata, so the
    extraction of the flavor from the instance, during notifications sent
    during unrescue, now succeed.

    Closes-Bug: #1277054
    Change-Id: I63bdd6c49505557829241869fd3efd2b2754530b