Using the near table would be most effective, as the Buffer could potentially find multiple features.
After you use generate the Near table, create a join between Poles and Lights using the ID for the nearest feature identified by the table. This will also allow you to filter out outliers and duplicates.