SCRaft: Achieving Fast and Stable Elections in Raft Consensus Algorithm

Existing strong leader consensus algorithms, such as the Raft algorithm and its prevote optimization, can encounter significant vote splitting rate and relatively long timeout periods during leader elections, resulting in prolonged election latency and extended service interruptions, especially in wide-area networks. This paper introduces SCRaft, a consensus algorithm that leverages the centralized advantages of a predecessor leader to reduce the latency associated with leader transitions in distributed systems. SCRaft is a re-invented Raft consensus algorithm that, by pre-designating a successive leader during the current leader’s term, which significantly speeds up the election process and minimizes service interruptions due to network packet loss or node failures. This approach not only minimizes disruptions due to network failures or node outages but also preserves the high throughput characteristic of the original Raft algorithm. We detail the algorithm design considerations and the critical aspects of its implementation. Theoretical proof formally demonstrates that SCRaft ensures log consistency, and empirical results confirm its efficacy in reducing election latency without compromising system throughput. Our algorithm provides a robust solution for enhancing the availability of distributed systems.