Graph Neural Networks (GNNs) are a powerful paradigm for modeling graph-structured data, but their reliance on iterative neighborhood aggregation makes training computationally expensive, especially on large graphs. To address the resulting scalability challenges, the database community has recently focused on optimizing GNN training from a data management perspective. This chapter summarizes the end-to-end GNN training pipeline along the graph data lifecycle, encompassing the graph preprocessing, the batch generation, the data transfer, and the model training stages, and reviews representative techniques that improve efficiency at both stage level and end-to-end pipeline level. We distinguish the challenges of static and dynamic graph settings, discussing efficient training strategies for each. Finally, we outline how the remainder of this book is organized around scalable GNN training methodologies.