Zab: High-performance broadcast for primary-backup systems
International Conference on Dependable Systems & Networks (DSN), 2011
https://ieeexplore.ieee.org/document/5958223
Zab: Zookeeper atomic broadcast，原子广播在分布式系统中也称为 total order broadcast。Zookeeper 使用 Zab 协议将系统的增量状态修改同步到备节点。
Due the dependence of an incremental state change on the sequence of changes previously generated, Zab must guarantee that if it delivers a given state change, then all other changes it depends upon must be delivered first.
ZooKeeper 对系统的要求是一个客户端的多个操作需要能同时执行，且严格按照操作的顺序提交。传统的 Paxos 实现没有直接提供这样的能力，当多个操作同时进行的时候，可能会出现下图的问题，虽然每个值都达成了共识，但不能保持操作之间的依赖关系。
Zab 进程有两个角色，leader 和 follower。
A leader concurrently executes the primary role and proposes transactions according to the order of broadcast calls of the primary.
Followers accept transactions according to the steps of the protocol.
Leader 也会执行 Follower 的步骤。
算法
需要持久化的数据：
Zab 协议分为三个阶段：
- discovery
- synchronization
- broadcast
其中 Synchronization 阶段 Leader 将 Discovery 阶段从 Follower 收集的 history 数据选择最新的数据（Step l.1.2）广播给 Follower 节点。文中提到这块可以做个优化:
use a leader election primitive that selects a leader that has the latest epoch and has accepted the transaction with highest zxid among a quorum of processes.
这个优化正是 Raft 的选举阶段的 Election restriction，Raft 将此作为一个 Safty Property。
Zab 协议的流程图示:
ZooKeeper’s Atomic Broadcast Protocol: Theory and practice 对 zab 原理及 Zookeeper 中 zab 的实现作了进一步介绍。
References
[1] Zab: High-performance broadcast for primary-backup systems by Adrian Colyer