sporadically in u/s CSIT jobs we are getting failures where instances are not able to
get their ip address. It looks like it might be that not all vxlan tunnels are getting
created.

in this failure it seems that compute #2 does not
have the tunnel to the control node. but, note that it does exist on the control node.
somewhere this creation is failing.

all logs under this directory

an identical passing job for comparison