In this paper, we consider the problem of joint user scheduling and dynamic pilot allocation in a Time-Division Duplex (TDD) based Massive MIMO network under varying traffic condition. One of the main problems with Massive MIMO systems is that the number of available orthogonal pilot signals is limited, and the dynamic allocation of these signals to different users is crucially needed to utilize the full benefit of these systems. In addition, pilot signals are radio resource control (RRC) configured in practice, and hence the frequent reconfiguration causes high signaling overhead and is costly. Using Lyapunov optimization framework, we develop an optimal algorithm that first assigns pilots dynamically based on queue sizes and the channel conditions of users as well as the reconfiguration cost at large time-scale. Then, it schedules users on a small-time scale. Numerical results show the efficacy of our algorithm and demonstrate that pilots do not need to be configured frequently at the expense of increased queue delay.