Graph Neural Networks (GNNs) have recently enabled substantial advances in graph learning. Despite their rich representational capacity, GNNs remain relatively under-explored for large-scale social modeling applications. One such industrially ubiquitous application in online social platforms is friend suggestion: platforms recommend their users other candidate users to befriend, to improve user connectivity, retention and engagement. However, modeling such user-user interactions on large-scale social platforms poses unique challenges: such graphs often have heavy-tailed degree distributions, where a significant fraction of users are inactive and have limited structural and engagement information. Moreover, users interact with different functionalities, communicate with diverse groups, and have multifaceted interaction patterns. We study the application of GNNs for friend suggestion, providing the first investigation of GNN design for this task, to our knowledge. To leverage the rich knowledge of heterogeneous in-platform user actions, we formulate friend suggestion as multi-faceted friend ranking with multi-modal user features and link communication features. We present a neural architecture, GraFRank, which is carefully designed to learn expressive user representations from multiple user feature modalities and user user interactions. Specifically, GraFRank handles heterogeneity in modality homophily via modality-specific neighbor aggregators, and learns non-linear modality correlations through cross-modality attention. We conduct experiments on two multi million user social network datasets from Snapchat, a leading and widely popular mobile social platform, where GraFRank outperforms several state-of-the-art approaches on candidate retrieval (by 30% MRR) and ranking (by 20% MRR) tasks. Moreover, our qualitative analysis suggests notable gains for critical populations of less-active and low-degree users.