Research on cross-ledger synchronization of real-time accounting system based on Spark Streaming and YARN scheduling

This paper proposes a real-time accounting system that integrates Spark Streaming (Spark Streaming -based stream execution) with YARN priority scheduling to mitigate task latency, resource imbalance, and cross-ledger inconsistency in large-scale streaming accounting workloads. A priority-aware scheduling framework is designed to dynamically allocate cluster resources according to queue pressure, task urgency, and runtime execution states. A deep reinforcement learning strategy is embedded to adaptively optimize task ordering and resource assignment under fluctuating streaming loads. For cross-ledger synchronization, a blockchain-enabled mechanism with smart contracts is implemented to ensure atomic commit, traceability, and consistency among distributed ledgers, while zero-knowledge proofs are employed to validate synchronization correctness with minimal disclosure of sensitive accounting fields. Task dependency relations are modeled using a graph neural network, and execution-time correlations are predicted via a long short-term memory network to support dependency-aware scheduling decisions. The system is evaluated on a enterprise-derived dataset containing 5000 real-time accounting instances. Experimental results show that the proposed method achieves a scheduling accuracy of 91.3% ± 0.5 and a synchronization accuracy of 88.0% ± 0.4, and reduces the average system response time to 1.7 s ± 0.1 under high-concurrency conditions. Ablation results further verify that reinforcement learning, graph-based dependency modeling, and blockchain-based cross-ledger coordination jointly contribute to the observed performance gains.