What should we pay attention to when implementing distributed locks in Redis? [Summary of precautions]

What should we pay attention to when implementing distributed locks in Redis? The following article will summarize and share with you some points to note when using Redis as a distributed lock. I hope it will be helpful to you!

Redis implements distributed locks

I recently saw an article while reading distributed locks Good article, specially processed for my own understanding:

Three core elements of Redis distributed lock implementation:

1. Locking

The easiest way is to use the setnx command. The key is the unique identifier of the lock, which is named according to the business. The value is the thread ID of the current thread. [Related recommendations:Redis Video Tutorial]

For example, if you want to lock the flash sale activity of a product, you can name the key "lock_sale_ID". And what is the value set to? We can temporarily set it to 1. The pseudo code for locking is as follows:

setnx(key, 1)When a thread executes setnx and returns 1, it means that the key does not originally exist and the thread successfully obtained it. Lock, when other threads execute setnx and return 0, it means that the key already exists and the thread failed to grab the lock.

2. Unlock

If you want to lock, you must unlock. When the thread that obtained the lock completes its task, it needs to release the lock so that other threads can enter. The simplest way to release the lock is to execute thedel instruction. The pseudo code is as follows:

del(key)After releasing the lock, other threads You can continue to execute the setnx command to obtain the lock.

3. Lock timeout

What does lock timeout mean? If a thread that obtains the lock dies while executing the task and has no time to explicitly release the lock, the resource will be locked forever, and other threads will never be able to come in again.

Therefore, the key of setnx must set a timeout period to ensure that even if it is not explicitly released, the lock will be automatically released after a certain period of time. setnx does not support timeout parameters, so additional instructions are needed. The pseudo code is as follows:

expire(key, 30)Taken together, the third step of our distributed lock implementation The first version of the pseudo code is as follows:

if（setnx（key，1） == 1）{ expire（key，30） try { do something ...... }catch()　　{　　}　　finally { del（key） } }

Because there are three fatal problems in the above pseudo code:

1.The non-atomicity of setnx and expire

Imagine an extreme scenario. When a thread executes setnx, it successfully obtains the lock:

setnx has just been successfully executed, and before it has time to execute the expire command, node 1 Duang hangs up. Lost.

if（setnx（key，1） == 1）{　　//此处挂掉了..... expire（key，30） try { do something ...... }catch() 　　{ 　　} 　　finally { del（key） } }

In this way, the lock does not have an expiration time set and becomes "immortal", and other threads can no longer obtain the lock.

How to solve it? The setnx instruction itself does not support the incoming timeout period.Redis 2.6.12 or above adds optional parameters to the set instruction. The pseudo code is as follows: set (key, 1, 30, NX), so that Can replace setnx instruction.

2. Usingdel after the timeout results in accidentally deleting the locks of other threads

Another extreme scenario, if a thread successfully obtains the lock, and the timeout is set to 30 seconds.

If for some reason thread A executes very slowly and has not finished executing after 30 seconds, the lock will be automatically released upon expiration and thread B will obtain the lock.

Subsequently, thread A completes the task, and thread A then executes the del instruction to release the lock. But at this time, thread B has not finished executing.Thread A actually deletes the lockadded by thread B.

How to avoid this situation? You can make a judgment before del releases the lock to verify whether the current lock is a lock added by yourself.

As for the specific implementation, you can use the current thread ID as the value when locking, and verify whether the value corresponding to the key is the ID of your own thread before deleting it.

加锁： String threadId = Thread.currentThread().getId() set（key，threadId ，30，NX） doSomething..... 解锁： if（threadId .equals(redisClient.get(key))）{ del(key) }

However, doing so implies a new problem,if judgment and lock release are two independent operations, not atomic.

We are all programmers who pursue the ultimate, so this part must be implemented using Lua script:

String luaScript = 'if redis .call('get', KEYS[1]) == ARGV[1] then returnredis.call('del', KEYS[1]) else return 0 end' ;

##redisClient.eval(luaScript, Collections.singletonList(key) , Collections.singletonList(threadId));

3.Possibility of concurrency

还是刚才第二点所描述的场景，虽然我们避免了线程A误删掉key的情况，但是同一时间有A，B两个线程在访问代码块，仍然是不完美的。

怎么办呢？我们可以让获得锁的线程开启一个守护线程，用来给快要过期的锁“续航”。

当过去了29秒，线程A还没执行完，这时候守护线程会执行expire指令，为这把锁“续命”20秒。守护线程从第29秒开始执行，每20秒执行一次。

当线程A执行完任务，会显式关掉守护线程。

另一种情况，如果节点1 忽然断电，由于线程A和守护线程在同一个进程，守护线程也会停下。这把锁到了超时的时候，没人给它续命，也就自动释放了。

memcache实现分布式锁

首页top 10, 由数据库加载到memcache缓存n分钟
微博中名人的content cache, 一旦不存在会大量请求不能命中并加载数据库
需要执行多个IO操作生成的数据存在cache中, 比如查询db多次
问题
在大并发的场合，当cache失效时，大量并发同时取不到cache，会同一瞬间去访问db并回设cache，可能会给系统带来潜在的超负荷风险。我们曾经在线上系统出现过类似故障。

解决方法

if (memcache.get(key) == null) { // 3 min timeout to avoid mutex holder crash if (memcache.add(key_mutex, 3 * 60 * 1000) == true) { value = db.get(key); memcache.set(key, value); memcache.delete(key_mutex); } else { sleep(50); retry()； } }

在load db之前先add一个mutex key, mutex key add成功之后再去做加载db, 如果add失败则sleep之后重试读取原cache数据。为了防止死锁，mutex key也需要设置过期时间。伪代码如下

Zookeeper实现分布式缓存

Zookeeper的数据存储结构就像一棵树，这棵树由节点组成，这种节点叫做Znode。

Znode分为四种类型：

• 1.持久节点（PERSISTENT）

默认的节点类型。创建节点的客户端与zookeeper断开连接后，该节点依旧存在 。

• 2.持久节点顺序节点（PERSISTENT_SEQUENTIAL）

所谓顺序节点，就是在创建节点时，Zookeeper根据创建的时间顺序给该节点名称进行编号：

• 3.临时节点（EPHEMERAL）

和持久节点相反，当创建节点的客户端与zookeeper断开连接后，临时节点会被删除：

• 4.临时顺序节点（EPHEMERAL_SEQUENTIAL）

顾名思义，临时顺序节点结合和临时节点和顺序节点的特点：在创建节点时，Zookeeper根据创建的时间顺序给该节点名称进行编号；当创建节点的客户端与zookeeper断开连接后，临时节点会被删除。

Zookeeper分布式锁恰恰应用了临时顺序节点。具体如何实现呢？让我们来看一看详细步骤：

• 获取锁

首先，在Zookeeper当中创建一个持久节点ParentLock。当第一个客户端想要获得锁时，需要在ParentLock这个节点下面创建一个临时顺序节点Lock1。

之后，Client1查找ParentLock下面所有的临时顺序节点并排序，判断自己所创建的节点Lock1是不是顺序最靠前的一个。如果是第一个节点，则成功获得锁。

这时候，如果再有一个客户端Client2前来获取锁，则在ParentLock下载再创建一个临时顺序节点Lock2。

Client2查找ParentLock下面所有的临时顺序节点并排序，判断自己所创建的节点Lock2是不是顺序最靠前的一个，结果发现节点Lock2并不是最小的。

So,Client2registersWatcherwith the node ranked only higher than itLock1for monitoringLock1Whether the node exists.This means thatClient2failed to grab the lock and entered the waiting state.

At this time, if another clientClient3comes to acquire the lock, download and create a temporary one inParentLockSequence NodeLock3.

Client3Find all the temporary sequence nodes belowParentLockand sort them to determine the node you createdLock3# Is ## the first one in the order? It turns out that nodeLock3is not the smallest.

Client3registersWatcherwith the nodeLock2that is ranked only higher than it, for monitoringLock2Whether the node exists. This means thatClient3also failed to grab the lock and entered the waiting state.

Client1gets the lock,Client2monitorsLock1,Client3listened toLock2. This just forms a waiting queue, much like theAQS (AbstractQueuedSynchronizer)thatReentrantLockrelies on in Java.

Release lock

Client1will display the instruction to call the delete nodeLock1.

Client1During the task execution, if Duang’s A crash will disconnect the Zookeeper server. According to the characteristics of the temporary node, the associated nodeLock1will be automatically deleted.

Client2has been monitoring the existence status ofLock1, when theLock1node is deleted,Client2will be notified immediately. At this time,Client2will query all nodes underParentLockagain to confirm whether the nodeLock2created by itself is the current smallest node. If it is the smallest, thenClient2naturally obtains the lock.

Client2also deletes nodeLock2due to task completion or node crash, thenCient3will be notified.

Client3successfully obtained the lock.

##For more programming-related knowledge, please visit:

! !

The above is the detailed content of What should we pay attention to when implementing distributed locks in Redis? [Summary of precautions]. For more information, please follow other related articles on the PHP Chinese website!