Consistent Distributed Reactive Programming with Retroactive Computation
To address the time-consistency challenge posed by multi-source asynchronous, delayed, or lossy data in IoT systems, this paper introduces a high-level programming abstraction supporting distributed reactive and retroactive joint computation. Methodologically, it proposes a novel typed core calculus that unifies reactive signals with retroactive computation—enabling consistent state recovery at arbitrary checkpoints without relying on a global clock. Its hybrid semantic model integrates an untyped object calculus with process calculus to uniformly formalize time-varying signals and historical value queries. The approach formally guarantees time consistency through rigorous type safety and semantic soundness proofs. Microbenchmark evaluations demonstrate low overhead and high feasibility of retroactive recovery under typical IoT workloads, including scenarios with packet loss and variable latency. Key contributions include: (i) the first typed calculus bridging reactive and retroactive paradigms for distributed IoT; (ii) a clock-free, checkpoint-agnostic consistency mechanism; and (iii) empirical validation of practicality in resource-constrained, unreliable network conditions.