屠雪真 陈小强

摘 要： 随着移动互联网、云计算等技术的发展，分布式系统以其易扩展、高可靠、灵活性强等优点成为了应用软件系统的首选架构。然而大型分布式系统的更新升级存在着过程复杂、时间长、新旧版本共存等问题。从研究分析分布式系统更新升级的特点和关键技术点出发，结合电信大型分布式系统实践中遇到的问题，提出了一种自动化的升级和数据迁移方法，采用逻辑顺序号保证数据的一致性，采用逻辑机架实施分区升级，设计了一种接力赛机制减少升级期间的数据迁移量，解决了分布式系统升级耗时长风险大的问题。实验结果表明，与现有的升级方式相比，分区升级方法缩短升级时间50%左右，将对业务的影响时长减小到秒级，提升了升级效率，并有效降低了升级风险。

关键词： 移动互联网; 云计算; 分布式系统; 高可靠; 版本升级

中图分类号： TP393

文献标志码： A

文章编号：1007-757X（2019）06-0042-05

Abstract： With the development of technologies such as mobile Internet and cloud computing， distributed systems have become the preferred architecture for application software systems because of their advantages of easy expansion， high reliability， and flexibility. However， the update and upgrade of large-scale distributed systems have problems such as complicated process， long time， and coexistence of new and old versions. Based on the analysis of the characteristics and key technologies of distributed system upgrade， and combined with the problems encountered in the practice of large-scale distributed telecom systems， in this paper， an automated upgrade and data migration method is proposed. The method uses logical sequence numbers to ensure data consistency， uses logical racks to implement partition upgrades. And a relay race mechanism is designed to reduce the amount of data migration during the upgrade. It solves the problem that the distributed system upgrade takes a long time and has a high risk. The experimental results show that compared with the existing upgrade method， the partition upgrade method shortens the upgrade time by about 50%， reduces the impact on the service to the second level， improves the upgrade efficiency， and effectively reduces the upgrade risk.

Key words： Mobile Internet; Cloud computing; Distributed system; High reliability; Version upgrade

0 引言

隨着移动互联网、社交网络、云计算等技术的快速发展，数据量呈爆炸式增长。对于日益增长的海量数据处理，许多应用面临着高并发、低时延、高可用及弹性扩展等挑战，分布式系统及相关技术被广泛地应用于各行各业的应用中。大型的分布式系统通常由几十到几百甚至几千台服务节点组成，构建和维护这种大规模分布式系统往往很复杂[1]。在系统的生命周期内，每隔一段时间，系统都需要进行升级以提升性能、修改错误或者增加新功能。……