中文摘要：

在大数据背景下,分布式系统被企业广泛部署和应用,随着分布式系统节点规模的扩大,系统故障的概率也将随之增加,在分布式系统中引入容错机制,对提升分布式系统可用性、可靠性、可恢复性至关重要.CLAIMS系统是面向金融领域的对实时数据进行实时分析的内存数据库系统——在数据不断注入系统时,提供近实时的查询、分析任务.本文主要探讨CLAIMS系统中容错机制.依据租约机制,实现系统中异常节点的快速发现及标记（即Fail-fast）.在标记异常节点之后,实现对受影响分析任务的重启（即Fail-over）;对异常节点全局内存状态的恢复（即Fail-back）.实验结果表明,本文所提算法能够较好地实现CLAIMS系统的容错特性.

英文摘要：

In the big data era, distributed system has been widely deployed and applied in various fields. Nevertheless, the more nodes involved, the higher probability of system failures may occur. It is important to introduce fault-tolerance mechanism for distributed systems to achieve even higher performance, higher reliability and higher availability. CLAIMS system is an in-memory database system for real-time data analysis, which is mainly used for financial applications. It provides near real time query task and analytic task. This paper mainly discuss fault-tolerance mechanism in CLAIMS. Achieve lease-based quick system failure detection （Fail-fast）. Achieve restart of affected analytic task after detecting failure （Fail-over）. Achieve in-memory state recovery of abnormal node. Experiment indicate that the algorithm presented in this paper can achieve fault-tolerance in CLAIMS.