Nuclear reaction rate databases serve as essential inputs for nucleosynthesis and stellar evolution modeling, directly influencing the accuracy and physical reliability of calculations in various nuclear astrophysics processes. This work comprehensively reviews the major reaction rate databases i.e. REACLIB, STARLIB, and BRUSLIB, highlighting their objectives, data structures, and representative applications, and discussing their coverage, fitting methods, and uncertainty evaluation. These databases have been instrumental in advancing the standardization of nuclear reaction network calculations. However, although these databases have significantly lowered the barrier to performing network modeling, there remains substantial room for improvement in aspects such as database unit structures, update mechanisms, and organizational frameworks. For example, detailed information on the underlying nuclear physics experiments or data analyses is often not included in REACLIB. Therefore, enhancing the stored metadata deserveds careful consideration, as it can significantly improve the reliability of astrophysical modeling. At the same time, the advancement of nuclear astrophysics reaction rate databases depends heavily on continuous progress at the experimental frontier. In recent years, innovative experimental techniques, such as novel 4π high-resolution detector arrays and γ-charged particle coincidence measurements, have been widely applied to studies of key nuclear astrophysics reactions, significantly expanding research capabilities. To meet the demands of cutting-edge astrophysical studies for accurate reaction rates, the real-time updating and systematic evaluation of experimental data for key reactions present both an important opportunity and an urgent challenge for the development of modern databases. Several important achievements of the JUNA Collaboration at the Jinping underground nuclear astrophysics facility are also presented in this paper, where low-background experiments are conducted. Compared with traditional extrapolations used in databases, these new low-energy measurements are found to provide more direct constraints on key reactions in nuclear astrophysics and crucial experimental support for continuously optimizing databases in the future.