filter_scheduler should select from a pool of best hosts, not the single best host

[John Herndon]

After executing the weighing functions, the filter_scheduler chooses the first host in the weighed_hosts list. However, it is very rarely the case in a large data center that a single host is "best". The code should randomly choose from a pool of n best hosts, where n is configurable.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/19494

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/19823

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/19823
Committed: http://github.com/openstack/nova/commit/8310be5b941931cb6b63a4d0f778cee27878095d
Submitter: Jenkins
Branch: master

commit 8310be5b941931cb6b63a4d0f778cee27878095d
Author: John Herndon <email address hidden>
Date: Wed Jan 16 16:56:44 2013 +0000

    filter_scheduler: Select from a subset of hosts.

    Adding a subset of best hosts to choose a host from, instead
    of choosing the host with the highest weight. The subset size is
    configurable. Setting the subset size to 1 will mimic the old
    behavior.

    DocImpact

    Change-Id: Icc6f8a7ca7cf992d99b872a45e241f13fa6e1536
    Fixes: bug 1098635