Currently it can happen that a tunnel is broken due to a network error but does not time out fast enough. In such cases the node might try to reestablish the tunnel and get redirected to a random other one.
It would be nice to be able to specify a flag that kills the old/other connection and takes over the tunnel.
This might also be somewhat related to the load-balancing feature.