I've been fighting this for roughly two days and I'm starting to think that possibly it's not my code but an interaction with the server.
I'm using test-kitchen[1] with the kitchen-vagrant[2] driver to spin up vagrant machines and run tests against them. I'm using Jenkins to run kitchen in containers in parallel.
Basically Jenkins runs a docker container with ruby + vagrant 1.9.2 and runs kitchen test all at the same time as another container with ruby + vagrant 1.9.1.
the container for vagrant 1.9.1 created a VM `vagrant-dynamic-1.9.1:
[vagrant-1.9.1] Bringing machine 'default' up with 'ovirt4' provider...
[vagrant-1.9.1] ==> default: Creating VM with the following settings...
[vagrant-1.9.1] ==> default: -- Name: dynamic-1.9.1
And the container for vagrant 1.9.2 (nearly the same time) created a VM `vagrant-dynamic-1.9.2`:
[vagrant-1.9.2] ==> default: Creating VM with the following settings...
[vagrant-1.9.2] ==> default: -- Name: dynamic-1.9.2
[vagrant-1.9.2] ==> default: -- Cluster: Default
If you look at the ss: 
the container 1.9.1 will wait for dynamic-1.9.1 and try to contact it at 192.168.2.54
the container 1.9.2 will wait for dynamic-1.9.2 and try to contact it at 192.168.2.55
But if you look at the logs, the 1.9.1 container started trying to work with 192.168.2.55 by creating a new key then talking to it:
[vagrant-1.9.1] default: Key inserted! Disconnecting and reconnecting using new SSH key...
[vagrant-1.9.1] Waiting for SSH service on 192.168.2.55:22, retrying in 3 seconds
Because 1.9.1 inserted a generated key into that box, the 1.9.2 container which _should_ be talking to it cannot now:
[vagrant-1.9.2] ==> default: Rsyncing folder: /home/jenkins/.kitchen/cache/ => /tmp/omnibus/cache
[vagrant-1.9.2] SSH authentication failed! This is typically caused by the public/private
[vagrant-1.9.2] keypair for the SSH user not being properly set on the guest VM. Please
[vagrant-1.9.2] verify that the guest VM is setup with the proper public key, and that
[vagrant-1.9.2] the private key path for Vagrant is setup properly as well.
Via the ruby sdk I create the VM and store the ID it responded with.
Then to get the IP:
server = env[:vms_service].vm_service(env[:machine].id)
nics_service = server.nics_service
nics = nics_service.list
ip_addr = nics.collect { |nic_attachment| env[:connection].follow_link(nic_attachment).reported_devices.collect { |dev| dev.ips.collect { |ip| ip.address if ip.version == 'v4' } } }.flatten.reject { |ip| ip.nil? }.first rescue nil
Given this code I can't think of any way that I would get the wrong IP unless somehow the server responded incorrectly, since the NIC's i've scanned and compiled across are tied directly to the server I created.
Any thoughts? This only happpens randomly and it seems to happen if I bombard the server with a bunch of VM creations simultaneously