Instance DHCP Request are Sometimes Answered by Remote Dnsmasq

[zk]

When an instance sends out a DHCPDISCOVER or DHCPREQUEST, it sometimes leaves the compute node and gets answered by Dnsmasq running on another compute node. The remote Dnsmasq will always respond with "DHCPNAK no address available" (to DHCPDISCOVER) or "DHCPNAK address not available" (to DHCPREQUEST), because it doesn't have an entry for that instance in its config file.

Syslog:
Jan 28 15:31:04 xxxxx dnsmasq-dhcp[10454]: DHCPREQUEST(brxxx) 192.168.0.x 12:34:56:78:90:ab
Jan 28 15:31:04 xxxxx dnsmasq-dhcp[10454]: DHCPNAK(brxxx) 192.168.0.x 12:34:56:78:90:ab address not available

Expected Behaviour:
According to blueprint (https://review.openstack.org/#/c/16578/), when share_dhcp_address is set to true, the dhcp messages should be firewalled using iptables and ebtables

Environment:
- Icehouse 2014.1.3
- Ubuntu 14.04
- Multihost mode
- Multiple compute nodes
- nova-network Vlan-Manager
- dnsmasq version 2.68
- nova-compute and nova-network run on the same node (other services run on other nodes)

[Sean Dague (sdague)]

This appears valid, would be good to produce a reproduce test

[Brent Eagles (beagles)]

Could this be a duplicate of: https://bugs.launchpad.net/nova/+bug/1318104 ?