Computing the importance scores of nodes in large-scale graphs through centrality measures enables the identification of important nodes while also supporting various real-life applications (e.g., preventing the dissemination of fake news). However, these graphs are often partitioned over several data centers, which are either mutually distrustful (e.g., owned by distinct entities possibly competing) or that fall under incompatible personal data protection laws (e.g., European and USA data centers). This prevents the centralization of the full graph on a single data center and thus directly hampers the computation of centrality measures. In this paper, we present our ongoing work on Poppy, our approach to tackle this problem. More precisely, Poppy consists of a suite of distributed algorithms dedicated to spectral centrality measures, running over the participating data centers, and leveraging the CKKS fully homomorphic encryption scheme. The performance of each Poppy variant depends on the graph structure and no Poppy algorithm is ultimately more efficient than others. In this short paper, we introduce the Poppy algorithms, discuss their costs as well as their correctness and security guarantees, position them with related work, and overview ongoing work on performance improvements, on the automatic selection of the best Poppy algorithm, and on the launch of experiments over large-scale real-life graphs.