MySQL 高可用架构.pptx
2016,MySQL 高可用架构,邵汉成,技术中心,CONTENTS,目录,需要达成的目标,高可用：冗余与备份,面向读操作的横向伸缩,面向写操作的横向伸缩,高大上的MySQL Cluster,Replication与BinLog,MySQL 监控,实现目标,高可用,响应快,可伸缩,易监控,百万订单/天 5百亿元/年,冗余与备份,冗余：当一个或几个Node出现问题时，系统可以正常运行 备份：使用备份可以将Node恢复到Crash之前的状态,快速响应,快速响应用户请求，用户体验好,可随时、按需Scale out,Scaling out reads Scaling out writes,监控系统的运行,Understand the load on our system Keep a keen eye on any problems,大容量 驴妈妈系统业务容量目标,订单量 100万单每天 销售量 500亿元每年,CONTENTS,目录,需要达成的目标,高可用：冗余与备份,面向读操作的横向伸缩,面向写操作的横向伸缩,高大上的MySQL Cluster,Replication与BinLog,MySQL 监控,高可用,提供冗余与备份,冗余：提供多台master和slave，防止单点故障或几个node故障 备份：Crash之后可以迅速恢复,冗余,故障发生时的恢复计划,Slave Failures Master Failures Disaster Recovery,计划,详细的执行方法,基本的准备工作：增加新Slaves，移除Slaves，切换Master Master Hot Standby Dual Masters Slave Promotion Circular Replication,方法,99.999%,高可用,Master Hot Standby 结构,高可用,Dual Masters 结构,Active Active: 两个Master同时接受写入，并将更新数据同步到另一个master。 常见用途：根据用户所在区域，用户访问就近的master Active Passive: 写操作写入active master，同时passive master只保持与active master 同步,高可用,Circular Replication 结构,All subscriber data is replicated to all sites, and updates of data are allowed in all data centers.,Slave Promotion,高可用,traditional method of slave promotion,revised method for promoting a slave,CONTENTS,目录,需要达成的目标,高可用：冗余与备份,面向读操作的横向伸缩,面向写操作的横向伸缩,高大上的MySQL Cluster,Replication与BinLog,MySQL 监控,Replication与BinLog机制,Replication基本介绍,Replication与BinLog机制,Binary Log 基本介绍,基于Statement的复制 基于Row的复制,Replication与BinLog机制,深入理解 Replication结构,Replication与BinLog机制,GTID: Global Transaction ID,Each transaction on a server is assigned a transaction identifier. assigned to a transaction based on the order in which they committed. The GTID of the transaction is retained regardless of the number of times that the transaction is propagated.,CONTENTS,目录,需要达成的目标,高可用：冗余与备份,面向读操作的Scale out,面向写操作的Scale out,高大上的MySQL Cluster,Replication与BinLog,MySQL 监控,读Scale out 对系统Load的帮助,面向读操作的Scale out,一台服务器容量为10000TPS，支持写操作4000TPS，读操作6000TPS：,增加3台Slaves之后：,同步或异步 Replication 的选择,面向读操作的Scale out,同步Replication的延迟和对TPS影响数据表,典型的拓扑图,面向读操作的Scale out,Hierarchical topology,路由与负载均衡,面向读操作的Scale out,Load balancing on the application level,Using a proxy to distribute queries,其他特性,面向读操作的Scale out,Filtering Replication Events Using Filtering to Partition Events to Slaves,CONTENTS,目录,需要达成的目标,高可用：冗余与备份,面向读操作的Scale out,面向写操作的Scale out,高大上的MySQL Cluster,Replication与BinLog,MySQL 监控,面向写操作的Scale out,Dual Masters 结构是否能Scale out 写操作？,面向写操作的Scale out,Sharding,Placing data geographically close to the user Reducing the size of the working set Distributing the work,Limitation of Sharding,Cross-shard joins Using AUTO_INCREMENT,面向写操作的Scale out,Sharding Key and Shard Mapping,根据Sharding Key决定如何将表分Shard List mapping Range mapping Hash mapping,面向写操作的Scale out,Processing Queries and Dispatching Transactions,How to dispatch transactions to the right shard How to get the sharding key for the transaction How to use caching to improve performance,CONTENTS,目录,需要达成的目标,高可用：冗余与备份,面向读操作的横向伸缩,面向写操作的横向伸缩,高大上的MySQL Cluster,Replication与BinLog,MySQL 监控,高大上的MySQL Cluster,MySQL Cluster 架构,CONTENTS,目录,需要达成的目标,高可用：冗余与备份,面向读操作的横向伸缩,面向写操作的横向伸缩,高大上的MySQL Cluster,Replication与BinLog,MySQL 监控,MySQL 监控,MySQL 监控,My SQL Monitoring Storage Engine Monitoring Replication Monitoring,THANK YOU,技术中心,
