Networks are natural geometric objects. Yet the discrete metric structure of shortest path lengths in a network, known as chemical distances, is definitely not the only reservoir of geometric distances that characterize many networks. The other forms of network-related geometries are the geometry of continuous latent spaces underlying many networks, and the effective geometry induced by dynamical processes in networks. A solid and rapidly growing amount of evidence shows that the three approaches are intimately related. Network geometry is immensely efficient in discovering hidden symmetries, such as scale-invariance, and other fundamental physical and mathematical properties of networks, as well as in a variety of practical applications, ranging from the understanding how the brain works, to routing in the Internet. Here, we review the most important theoretical and practical developments in network geometry in the last two decades, and offer perspectives on future research directions and challenges in this novel frontier in the study of complexity.