We address the problem of distribution network reconfiguration (DNR) under uncertainty, formulated as a mixed-integer second-order cone program (MISOCP). The formulation accurately represents power flow physics together with discrete switching and on-load tap changer (OLTC) positions whose settings are determined over a mid-term planning horizon. Uncertainty in load and generation during this horizon is represented through multiple operating scenarios. However, its computational complexity increases rapidly with network size and scenarios, making standard solvers impractical for large or real-world urban distribution systems with many scenarios. To overcome this challenge, we propose a scalable, consensus-based algorithm that integrates the MISOCP formulation with the alternating direction method of multipliers (ADMM). The algorithm decomposes the DNR problem across multiple scenarios, allowing subproblems to be solved independently while maintaining global consistency in discrete switching and OLTC decisions. A loop-based switch grouping strategy reduces combinatorial complexity, while an adaptive penalty update enhances convergence speed and numerical stability. The proposed approach is validated on the IEEE 33-node benchmark system, as well as on a real urban distribution network in city of Lausanne, Switzerland, with 1004 nodes, including a simplified 234-node version. Results show that the proposed algorithm efficiently identifies feasible radial topologies that satisfy all voltage and current constraints.
