When viewing your cluster nodes you may notice these errors when viewing replication status.
As a result replication will be broken and make continue to break even after re-initializing.
Last_Error: Error 'Duplicate entry '05441993516f45ec9b5d710a5103e625-\x86\xF0\x82\xD4\xF7\xFA\xB5\xA' for key 'nodeid'' on query. Default database: 'ssg'. Query: 'insert into service_metrics (nodeid, end_time, interval_size, back_max, front_max, back_min, front_min, attempted, authorized, completed, period_start, resolution, published_service_goid, service_state, start_time, back_sum, front_sum, goid) values ('05441993516f45ec9b5d710a5103e625', 1528452847862, 5000, 498, 499, 202, 203, 2, 2, 2, 1528452845000, 0, x'86F082D4F7FAB5A82B13E817B6C5EC9A', 'ENABLED', 1528452847284, 700, 702, x'683970D16A5160A03DDE48318846D5F8')'
API Gateway 9.4
You also want to make sure that Primary and secondary using same primary DB
there was number of cases where secondary was configured to use its own DB while primary was using its own .. so replication was being broken due to two active DBs ( duplicate entries ...)