This scenario is checked in icehouse release.
Build used:
===========
neutron-common : 1:2014.1-0ubuntu1~cloud0
neutron-dhcp-agent : 1:2014.1-0ubuntu1~cloud0
neutron-l3-agent : 1:2014.1-0ubuntu1~cloud0
neutron-metadata-agent :1:2014.1-0ubuntu1~cloud0
neutron-plugin-ml2 :1:2014.1-0ubuntu1~cloud0
neutron-plugin-nicira : 1:2014.1-0ubuntu1~cloud0
neutron-plugin-vmware : 1:2014.1-0ubuntu1~cloud0
neutron-server: 1:2014.1-0ubuntu1~cloud0
python-neutron: 1:2014.1-0ubuntu1~cloud0
python-neutronclient : 2:2.3.4.46.g07bcee8+git201404070301~trusty-0ubuntu1
Steps performed:
================
1. Through network horizon, create one network with one subnet. name of this subnet is "subnet1" and his network address is 85.85.85.0/24
2. attached this subnet to router's interface.
3. Through horizon or nova command create 2 to 3 VMs/instances
All these Vm are correctly created and connected to subnet1. They are getting correct IP address from subnet1.
4. Now through horizon, create second subnet. His name is "subnet2" and his network address is 95.95.95.0/24.
5. Through horizon or nova command create 2 VMs/instances and check they are connecting to which subnet networks.
Observed behavior:
=================
1. While creating the VM, we can not add these VM to second subnet of network.
2. It will always add to first subnet only.
3. there is no option for this in both openstack horizon and nova command. If there are multiple subsets for same networks then there is no option to specify particular subnet while creating a instance.
4. All created VMs are automatically attached to first subnet only.
Using the Nova CLI you can pre-create the port with a specific Fixed IP and then use the port uuid in the nova boot command. That said, it always baffled me the use case where a tenant network (that models a single L2 broadcast domain) needs to be mapped by two logical IP network subnets.