The test_reassign_port_between_servers test in Tempest creates
a port in neutron and two servers. It attaches the port to the
first server and then quickly detaches it and waits for the
port.device_id to be unbound from the server. Then it repeats
that for the second server.
The interface detach from the guest is asynchronous and happens
before nova unbinds the port, so there is a race where the port's
device_id is unset but the interface is still on the first guest
when we try to attach to the second guest, which fails.
This is a latent bug, but apparently has been tickled by the
move to our neutron CI jobs to use ubuntu xenial nodes.
The fix is to add a detach and retry loop on the interface detach
on the guest so that we can wait until the interface is gone
from the guest before nova unbinds the port in neutron, which is
what the Tempest test is waiting for. Then the device should be
available for attaching to the second guest.
This is similar to what we do with detaching volumes.
Closes-Bug: #1607714
Change-Id: Ic04aad8923ea2edf1d16e32c208cd41fdf898834
a3b3e8d831