Social network alignment, which aims to uncover corresponding users within online social networks, is the prerequisite for various cross-network tasks. Most existing methods adhere to relation completeness and consistency for un-directed graphs. Nevertheless, due to diverse functionalities, platform disparity is common across universally incomplete online social networks, where directed relations like follower-followee and subgraph isomorphism from indistinguishable neighborhoods are ubiquitous. Formulating online social networks as directed and attributed, we propose a framework, GRAME, to jointly alleviate relation incompleteness, platform disparity, and subgraph isomorphism. For integrated mitigation, we successively develop relation augmentation on pre-alignments, design multi-view embeddings for less platform disparity, and devise similarity refinement aware of subgraph isomorphism. Integrating refined similarities, we match users iteratively in gradual alignment for better performances. We further develop an unsupervised variant, UGRAME, with the two-stage prior alignment generation. Extensive experiments on real-world datasets demonstrate that both GRAME and UGRAME outperform thirteen state-of-the-art methods.