Ironic: Unavailable nodes may be scheduled to

[Jim Rollenhagen]

The Ironic driver reports all resources consumed for compute nodes in certain unavailable states (e.g. deploying, cleaning, deleting). However, if there is not an instance associated with the node, the resource tracker will try to correct the driver and expose these resources. This may result in being scheduled to a node that is still cleaning up from a previous instance.

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/188652
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=f0ca0f73b131838f1441ff4affae10fe78ff7150
Submitter: Jenkins
Branch: master

commit f0ca0f73b131838f1441ff4affae10fe78ff7150
Author: Jim Rollenhagen <email address hidden>
Date: Thu Jun 4 19:22:15 2015 -0700

    Ironic: Don't report resources for nodes without instances

    Ironic falsely reports resources as consumed for certain states. Some
    of these states are not associated with an instance, and the resource
    tracker will notice that and "correct" the driver. This allows nodes
    in CLEANING, etc. to be scheduled to.

    Fix this by only reporting resources consumed when an instance UUID
    is associated with the node. This association happens before the deploy
    starts, and is removed after Nova sees tear down as complete, so it
    should be safe to only look at the instance UUID for this. For any
    "unavailable" state, report that there are zero resources available,
    never allowing the node to be scheduled to.

    Closes-Bug: #1462374
    Change-Id: I48913f724a1bbe3711ea543a50b9ba096d5e95d7