Nested Pure Operation-Based CRDTs

Modern distributed applications increasingly replicate data to guarantee high availability and optimal user experience. Conflict-free Replicated Data Types (CRDTs) are a family of data types specially designed for highly available systems that guarantee some form of eventual consistency. Designing CRDTs is very difficult because it requires devising designs that guarantee convergence in the presence of conflicting operations. Even though design patterns and structured frameworks have emerged to aid developers with this problem, they mostly focus on statically structured data; nesting and dynamically changing the structure of a CRDT remains to be an open issue.

This paper explores support for nested CRDTs in a structured and systematic way. To this end, we define an approach for building nested CRDTs based on the work of pure operation-based CRDTs, resulting in nested pure operation-based CRDTs. We add constructs to control the nesting of CRDTs into a pure operation-based CRDT framework and show how several well-known CRDT designs can be defined in our framework. We provide an implementation of nested pure operation-based CRDTs as an extension to the Flec, an existing TypeScript-based framework for pure operation-based CRDTs. We validate our approach, 1) by implementing a portfolio of nested data structures, 2) by implementing and verifying our approach in the VeriFx language, and 3) by implementing a real-world application scenario and comparing its network usage against an implementation in the closest related work, Automerge. We show that the framework is general enough to nest well-known CRDT designs like maps and lists, and its performance in terms of network traffic is comparable to the state of the art.