Dhcp db interaction is horribly inefficient

[Vish Ishaya]

In linux_net:get_dhcp_leases and get_dhcp_hosts two db requests are made for every running instance in the db. The implication is that 2 * instances queries are made every time an instance is launched or terminated.

Ultimately this code needs to be refactored to just send in the relevant information to linux_net The information seems to be:
network_id
instance_hostname
instance_host
fixed_ip_address
vif_address
instance_id (could probably use hostname in place of this)

This is too large of a change for essex though so I think we need something like the following:

a) use joins to load the instance and vif in the db.network_get_associated_fixed_ips query (only used in linux_net)

b) optionally pass in a host parameter to filter the query

[Vish Ishaya (vishvananda)]

The included patch is what the interface should look like. Underlying query needs to be written.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/5518

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/5518
Committed: http://github.com/openstack/nova/commit/ffb9361596c1708d7d009c195b4b3bc74d00a01c
Submitter: Jenkins
Branch: master

commit ffb9361596c1708d7d009c195b4b3bc74d00a01c
Author: Vishvananda Ishaya <email address hidden>
Date: Mon Mar 19 09:01:16 2012 -0700

    Improve performance of generating dhcp leases

     * Dont make 2 * instances queries when generating dhcp leases
     * Uses some nasty joins to minimize necessary changes
     * Adds test to verify functionality
     * Uses old sqlalchemy compatible join syntax
     * Uses vif_id as nw identifier for single_host_gateway mode
     * Cleans up tests in linux_net
     * Fixes bug 959378

    Change-Id: I1c74d114dd7caad1781180585bb65691e536bcf9