Time of "no capabilities" after nova-scheduler starts

[Arata Notsu]

Right after a nova-scheduler starts, the scheduler doesn't know capabilities of services at all (i.e. HostManager.service_states is empty). Since capabilities are sent by periodic_task, it takes some time for the scheduler to get to know the capabilities. If a "run_instance" request comes in this duration, the scheduler cannot route the request to suitable nova-compute.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/14819

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/14819
Committed: http://github.com/openstack/nova/commit/9ac10f55e4b40ac496c20f56501e5aeffa196692
Submitter: Jenkins
Branch: master

commit 9ac10f55e4b40ac496c20f56501e5aeffa196692
Author: Arata Notsu <email address hidden>
Date: Thu Oct 25 20:24:23 2012 +0900

    Let scheduler know services' capabilities at startup

    Fixes bug 1071254.

    Changes:
    * Add new rpc-api(fanout) of compute "publish_service_capabilities"
        This rpc-api urges services to send its capabilites to the scheduler.
    * Scheduler calls publish_service_capabilities right after the start

    By them, the scheduler get to know the capabilities earlier.

    Now we can expect that the scheduler always holds the capabilities. So it
    is reasonable to change HostManager to ignore hosts whose capabilities are
    "None" since it becomes a rare case; this will make scheduling more
    reliable. This will achieved by Another patch.

    Change-Id: If6582765011fd5e1b794bfdc068e17630ba381cb