Recovery from failure of a working path for communications traffic in a connection-oriented network, where the working path has a preplanned recovery path, involves initiating recovery by requesting computation of a new recovery path for the communications traffic to avoid the failure. If the recovery is not successful within a time limit, recovery is carried out using the preplanned recovery path. Determining if the recovery is not successful and initiating the recovery using the preplanned recovery path can be controlled by an ingress node of the working path. By trying to compute a new recovery path first, network resources can be used more efficiently, as the new recovery path is likely to reuse most of the working path, since it is computed knowing the location of the fault.