Hybrid beamforming is a promising technology for the millimeter-wave Massive MIMO communication system towards practicality. The hybrid architecture makes user scheduling face new challenges and becomes a joint optimization of analog beam selection and user selection. This paper considers the practical scenario that the transmitter can not obtain perfect channel state information (CSI), the beam training based on fixed codebook is used to obtain equivalent CSI. Through introducing scheduling indicator function and effective approximations, the joint optimization problem is modeled into non-convex combination program. To address this program, a mathematical method based on difference of two convex functions (DC) and branch-and-bound (BB) technique is proposed. The simulation results show that DCA & BB method can achieve an average sum rate close to exhaustive searching.