Over the last decade, building on advances in the areas of knowledge compilation and weighted model counting has drastically increased the scalability of inference in probabilistic logic programs. In this paper, we provide an overview of how this has been possible and point out some open challenges.