Network devices can use maximally redundant trees (MRTs) for delivering traffic streams across a network, and for transitioning traffic to a new set of MRTs after a topology change, without dropping traffic. The disclosure describes distributed computation of a set of MRTs from one or more ingress devices to one or more egress devices of the network. In one example, network devices in a network compute a set of MRTs, and establish a set of LSPs along the paths of the set of MRTs. After a change to the network topology, convergence sequencing is managed by a central controller, which centrally orchestrates the sequence for moving traffic from being sent on the old MRT paths to being sent on newly computed MRT paths after the controller determines that all new MRT forwarding state has been installed on the network devices.