ZooKeeper+: The Optimization of Election Algorithm in Complex Network Circumstance

摘要

Dynamic configuration management brings challenge for the distributed file systems while keeping the normal service. In this paper, we describe a robust election algorithm based on ZooKeeper, we realize the dynamic addition and deletion of servers without service interruption. There is only one clustered mode for servers without switching between prior two modes, it also speeds up the leader election. The leader maintains an active server list which speeds up handling of the transaction. The algorithm also ensures the data consistency and system stability against all possible issues. Through the evaluation, it takes not much overhead to realize the addition and deletion of servers and the recovery of crashed servers under various complex network circumstances, and it takes little more election time and initialization time of service to obtain the greater scalability.