Why you want to replicate data?

• 保持数据在地理上更接近用户来降低延迟
• 使系统在一部分节点崩溃时，依然可以继续使用，提高系统可用性
• 增加可以用于读操作的节点，提高数据吞吐量

Three popular algorithms for replicating changes between nodes

• single leader
• multi-leader
• leaderless

Leaders and Followers

Synchronous vs Asynchronous

innobackupexMySQL设置快照工具

binlog coordinates snapshot在MySQL日志中的位置

Implementation of Replication Logs

• Statement-based replication
• Write-ahead log(WAL) skipping
• Logical(row-based) log replication
• Trigger-based replication

Problems with Replication Lag

因为同步延迟造成的数据不一致性，往往会因为集群规模的扩大而愈发凸显（集群越大，网络情况也越发复杂）。