Various techniques can be used to avoid loops during network convergence after a topology change such as a failure. For example, a method can involve detecting a failure that disrupts an existing forwarding path between a node and a destination node; calculating at least one updated forwarding path from the node to the destination node; identifying a maximum time for at least a portion of the network to reach a post-convergence state with respect to the failure; until the maximum time has elapsed at the node, explicitly specifying at least a portion of the updated forwarding path in packet headers of any packets being sent from the node to the destination node; and after the maximum time has elapsed at the node, non-explicitly specifying the portion of the updated forwarding path in packet headers of any additional packets being sent from the node to the destination node.