The Reconfigurable Intelligent Surface (RIS) introduces a new communication paradigm for the next generation of mobile communication systems, leveraging its programmable capabilities to shape the wireless electromagnetic environment. To further enhance the array gain of the RIS and increase capacity of the system, the RIS evolves to the extremely large-scale RIS (XL-RIS), which presents significant challenges for channel estimation. However, beam training stands as an effective way to acquire accurate channel state information (CSI). In this paper, we present a near-field beam training method for millimeter-wave (mmWave) wireless communication systems assisted the XL-RIS. Specifically, we firstly design the near-field codebook based on the near-field channel of the XL-RIS. Then, we propose the alternative hierarchical beam training (AHBT) scheme, where angle and distance parameters are alternately determined by using the classic hierarchical algorithm. Simulation results demonstrate that the proposed scheme achieves a performance of 94.34% with less than 1% overhead compared to exhaustive search method.