【摘要】 设计并实现了面向对象的分布式文件系统元数据服务器高可用方案,用于提高存储系统的可用性.系统使用集中式元数据管理服务器,通过日志文件和检查点文件对元数据进行保存;针对系统特点,该方案采用active/hot-standby模式实现元数据服务器冗余备份.对系统状态监控、日志及检查点数据同步复制、元数据服务器节点失败接管、防止系统split-brain等关键技术问题进行了深入研究和提出相应解决方法,并对影响系统恢复时间的因素进行了细致分析.测试表明,高可用功能的实现对系统性能影响可以随存储文件的增大而减少,并可在失败发生后的较短时间内完成主从服务器的切换.更多还原

【Abstract】 Design and implemented the Metadata High Availability in object-oriented Distributed File System to provide high availability to the whole storage system.The system uses a centralized metadata server and stores the metadata in log and checkpoint files.In the light of the system characters,redundant metadata server is used with active/hot-standby model.The key technical problems such as monitoring server status,replicating checkpoint and journal synchronously,failing over the failed server node and avoiding the occurrence of split-brain are deeply researched and corresponding solutions are provided.The factors that influence the recovery time are analyzed in detail.The test results showed that such distributed file systems can recovery from a metadata server failure in a short time with only a tiny performance influence when storing moderate size files.更多还原