nova-dhcpbridge shouldn't send notifications on old leases

[Vish Ishaya]

It is a serious performance issue with a large number of instances for nova-dhcpbridge to send lease updates for old leases. This results in one message per vm every time an instance is launched or terminated.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/5510

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/5510
Committed: http://github.com/openstack/nova/commit/4071161f3860dbceb091913d46adbb804d9c6893
Submitter: Jenkins
Branch: master

commit 4071161f3860dbceb091913d46adbb804d9c6893
Author: Vishvananda Ishaya <email address hidden>
Date: Mon Mar 19 07:56:07 2012 -0700

    Stop notifications from old leases

     * Fixes performance issue with lots of instances
     * Also removes some commented out code
     * Fixes bug 959356

    Change-Id: I6e4cf3cd6ebe9d5d1ae4fc2472ef1765d13f49f6

[Bertrand Lallau (bertrand-lallau)]

Hi,

I am fully agree with the serious performance issue concerning the network node in case of large number of instances.
Nevertheless ignoring "old" dhcp lease from dnsmasq seems to be the wrong way to correct this issue, and is dangerous.
In case of renew DHCP request (DHCPREQUEST with src IP address) Dnsmasq never launch nova-dhcpbridge script with "add" argument but with "old" one.
Hence lease_fixed_ip will not be run and fixed_ip "updated_at" and "lease" fields will never be updated in database.
Consequently the periodic task calling "fixed_ip_disassociate_all_by_timeout" can disassociate the fixed ip even if DHCP renew was sent by the VM and so fresh disassociate IP can be reuse badly for another VM.