network_get_all_by_host query fails to use index

[Vish Ishaya]

The get_networks_by_host query fails to use the existing indexes on fixed ips. This means a large number of fixed_ips can lead to the database getting bogged down and taking multiple seconds to return.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/121276

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/121276
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=64b2aa71d90f15bc5f0a27ae6534515f2ac1309d
Submitter: Jenkins
Branch: master

commit 64b2aa71d90f15bc5f0a27ae6534515f2ac1309d
Author: Vishvananda Ishaya <email address hidden>
Date: Fri Sep 12 17:29:42 2014 -0700

    Fixes network_get_all_by_host to use indexes

    The OR in the subquery was constructed in a way that didn't allow
    the fixed_ips without instances allocated to be filtered out. It
    also had an unnecessary join through virtual interfaces. We have a
    denormalized field for instance_uuid with an index in the fixed_ips
    table that we can use. This patch removes the unnecessary join and
    ads an extra filter inside the OR to reject fixed_ips without
    instances. In my test with 300000 fixed ips, this reduced the query
    time from about 2.5 seconds to less than 0.01 seconds.

    Change-Id: I26fb0dfaf5ff64bd70c947982de4ec1728c84856
    Resolves-bug: #1369009