分布式锁三种实现方式

Reference

[1] https://redis.io/topics/distlock

[2] https://dzone.com/articles/distributed-lock-using

[3] https://en.wikipedia.org/wiki/Distributed_lock_manager

Java的lock只作用于同一个JVM，当有多个instance同时处理requests时，java的锁已不再适用，所以分布式锁已成为必须。

目前分布式锁的实现方式主要有三种

1. 基于数据库(不在内存中)
2. 基于Zookeeper
3. 基于Redis(in memory)

一 基于数据库

我到真是见过一个组用sybase实现分布式锁。具体方式是针对每一个要加锁的object在数据库里建一个entry，表示某个instance正在占有这个object。原理很简单，只需要一个单独的LockMaintenanceService独立于应用之外负责管理。但是具体performance跟数据库相关，一般来说效率不会很好。

二 基于Zookeeper

ZooKeeper is a centralized service for maintaining configuration information, naming, providing distributed synchronization, and providing group services. 

用Zookeeper做分布式锁也很常见，具体方式：

For a single task say its id is 10000, lib-zookeeper will create a parent node (with random bytes) in Zookeeper called "task-10000". Suppose "thread-1-1" comses from instance 1, and is trying to get the lock on node "task-10000", lib-zookeeper will create a child node like "task-10000/***sessionId*thread-1-1***" under the parent node and give the lock to thread-1-1. A watcher is created and will be updated each time the zookeeper lock gets released. 

So the next time thread-2-1 from instance 2 comes in and trying to acquire the lock, a new child node is created as "task-10000/***sessionId*thread-2-1***" under this parent node. If the lock is not released yet, it will just wait. If thread-1-1 finally releases the lock, the watcher will be notified and then the lock will be assigned to thread-2-1. 

Use try lock with timeout to avoid dead lock.

三 基于Redis

Redis is an open source (BSD licensed), in-memory data structure store, used as a database, cache and message broker. 

Redis比平常数据库的performance要好一些，毕竟in memory。可以用redis实现基于Redlock Algorithm的分布式锁。

四 总结

1. 基于分布式的锁一定要auto release，或者try-lock mechanism避免客户端crash或者死锁

2. 一个instance占用分布式锁的时间不应太长，否则instance越多效率越低，吞吐量越小，timeout越多