Comment 9 for bug 1737465

Reviewed: https://review.openstack.org/569218
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=a1a335f19a986a593487a802d7b9448f7efc5bab
Submitter: Zuul
Branch: master

commit a1a335f19a986a593487a802d7b9448f7efc5bab
Author: Matt Riedemann <email address hidden>
Date: Thu May 17 14:36:23 2018 -0400

    Avoid unnecessary joins in HostManager._get_instances_by_host

    While getting a HostState object for a given compute node during
    scheduling, if the HostState does not have its instance info
    set, either because it's out of date or because config option
    "track_instance_changes" is False, the HostManager still pulls
    the list of instances for that host from the database and stores
    it in HostState.instances.

    This is *only* used (in-tree) by the affinity filters and even
    then the only thing those filters use from HostState.instances
    is the set of keys from the dict, which is the list of instance
    UUIDs on a given host. The actual Instance objects aren't used
    at all. See blueprint put-host-manager-instance-info-on-a-diet
    for more details on that.

    The point of this change, is that when we go to pull the set
    of instances from the database for a given host, we don't need
    to join on the default columns (info_cache and security_groups)
    defined in the _instance_get_all_query() method in the DB API.

    This should be at least some minor optimization in scheduling
    for hosts that have several instances on them in a large cloud.
    As noted in the comment in the code, any out of tree filters
    that rely on using the info_cache or security_groups from the
    instance are now going to be hit with a lazy-load penalty
    per instance, but we have no contract on out of tree filters
    so if this happens, the people maintaining said filters can
    (1) live with it (2) fork the HostManager code or (3) upstream
    their filter so it's in-tree.

    A more impactful change would be to refactor
    HostManager._get_host_states to bulk query the instances on
    the given set of compute nodes in a single query per cell. But
    that is left for a later change.

    Change-Id: Iccefbfdfa578515a004ef6ac718bac1a49d5c5fd
    Partial-Bug: #1737465