亲宝软件园·资讯

展开

【分布式锁】04-使用Redisson实现ReadWriteLock原理

一枝花算不算浪漫 人气:2

前言

关于读写锁,大家应该都了解JDK中的ReadWriteLock, 当然Redisson也有读写锁的实现。

所谓读写锁,就是多个客户端同时加读锁,是不会互斥的,多个客户端可以同时加这个读锁,读锁和读锁是不互斥的

Redisson中使用RedissonReadWriteLock来实现读写锁,它是RReadWriteLock的子类,具体实现读写锁的类分别是:RedissonReadLockRedissonWriteLock

Redisson读写锁使用例子

还是从官方文档中找的使用案例:

RReadWriteLock rwlock = redisson.getReadWriteLock("tryLock");

RLock lock = rwlock.readLock();
// or
RLock lock = rwlock.writeLock();

// traditional lock method
lock.lock();

// or acquire lock and automatically unlock it after 10 seconds
lock.lock(10, TimeUnit.SECONDS);

// or wait for lock aquisition up to 100 seconds 
// and automatically unlock it after 10 seconds
boolean res = lock.tryLock(100, 10, TimeUnit.SECONDS);
if (res) {
   try {
     ...
   } finally {
       lock.unlock();
   }
}

Redisson加读锁逻辑原理

public class RedissonReadLock extends RedissonLock implements RLock {
    @Override
    <T> RFuture<T> tryLockInnerAsync(long leaseTime, TimeUnit unit, long threadId, RedisStrictCommand<T> command) {
        internalLockLeaseTime = unit.toMillis(leaseTime);

        return commandExecutor.evalWriteAsync(getName(), LongCodec.INSTANCE, command,
                                "local mode = redis.call('hget', KEYS[1], 'mode'); " +
                                "if (mode == false) then " +
                                  "redis.call('hset', KEYS[1], 'mode', 'read'); " +
                                  "redis.call('hset', KEYS[1], ARGV[2], 1); " +
                                  "redis.call('set', KEYS[2] .. ':1', 1); " +
                                  "redis.call('pexpire', KEYS[2] .. ':1', ARGV[1]); " +
                                  "redis.call('pexpire', KEYS[1], ARGV[1]); " +
                                  "return nil; " +
                                "end; " +
                                "if (mode == 'read') or (mode == 'write' and redis.call('hexists', KEYS[1], ARGV[3]) == 1) then " +
                                  "local ind = redis.call('hincrby', KEYS[1], ARGV[2], 1); " + 
                                  "local key = KEYS[2] .. ':' .. ind;" +
                                  "redis.call('set', key, 1); " +
                                  "redis.call('pexpire', key, ARGV[1]); " +
                                  "redis.call('pexpire', KEYS[1], ARGV[1]); " +
                                  "return nil; " +
                                "end;" +
                                "return redis.call('pttl', KEYS[1]);",
                        Arrays.<Object>asList(getName(), getReadWriteTimeoutNamePrefix(threadId)), 
                        internalLockLeaseTime, getLockName(threadId), getWriteLockName(threadId));
    }
}

客户端A(UUID_01:threadId_01)来加读锁

注:
以下文章中客户端A用:UUID_01:threadId_01标识
客户端B用:UUID_02:threadId_02标识

KEYS:

  • KEYS1: getName() = tryLock
  • KEYS[2]: getReadWriteTimeoutNamePrefix(threadId) = {anyLock}:UUID_01:threadId_01:rwlock_timeout

ARGV:

  • ARGV1: internalLockLeaseTime = 30000毫秒
  • ARGV[2]: getLockName(threadId) = UUID_01:threadId_01
  • ARGV[3]: getWriteLockName(threadId) = UUID_01:threadId_01:write

接着对代码中lua脚本一行行解读:

  1. hget anyLock mode 第一次加锁时是空的
  2. mode = false,进入if逻辑
  3. hset anyLock UUID_01:threadId_01 1
    anyLock是hash结构,设置hash的key、value
  4. set {anyLock}:UUID_01:threadId_01:rwlock_timeout:1 1
    设置一个string类型的key value数据
  5. pexpire {anyLock}:UUID_01:threadId_01:rwlock_timeout:1 30000
    设置key value的过期时间
  6. pexpire anyLock 30000
    设置anyLock的过期时间

此时redis中存在的数据结构为:

anyLock: {
  "mode": "read",
  "UUID_01:threadId_01": 1
}

{anyLock}:UUID_01:threadId_01:rwlock_timeout:1  1
客户端A 第二次来加读锁

继续分析,客户端A已经加过读锁,此时如果继续加读锁会怎样处理呢?

  1. hget anyLock mode 此时mode=read,会进入第二个if判断
  2. hincrby anyLock UUID_01:threadId_01 1 此时hash中的value会加1,变成2
  3. set {anyLock}:UUID_01:threadId_01:rwlock_timeout:2 1
    ind 为hincrby结果,hincrby返回是2
  4. pexpire anyLock 30000
  5. pexpire {anyLock}:UUID_01:threadId_01:rwlock_timeout:2 30000

此时redis中存在的数据结构为:

anyLock: {
  “mode”: “read”,
  “UUID_01:threadId_01”: 2
}

{anyLock}:UUID_01:threadId_01:rwlock_timeout:1  1
{anyLock}:UUID_01:threadId_01:rwlock_timeout:2  1

客户端B (UUID_02:threadId_02)第一次来加读锁

基本步骤和上面一直,加锁后redis中数据为:

anyLock: {
  "mode": "read",
  "UUID_01:threadId_01": 2,
  "UUID_02:threadId_02": 1
}

{anyLock}:UUID_01:threadId_01:rwlock_timeout:1  1
{anyLock}:UUID_01:threadId_01:rwlock_timeout:2  1
{anyLock}:UUID_02:threadId_02:rwlock_timeout:1  1

这里需要注意一下:
为哈希表 key 中的域 field 的值加上增量 increment,如果 key 不存在,一个新的哈希表被创建并执行 HINCRBY 命令。

Redisson加写锁逻辑原理

Redisson中由RedissonWriteLock 来实现写锁,我们看下写锁的核心逻辑:

public class RedissonWriteLock extends RedissonLock implements RLock {
    @Override
    <T> RFuture<T> tryLockInnerAsync(long leaseTime, TimeUnit unit, long threadId, RedisStrictCommand<T> command) {
        internalLockLeaseTime = unit.toMillis(leaseTime);

        return commandExecutor.evalWriteAsync(getName(), LongCodec.INSTANCE, command,
                            "local mode = redis.call('hget', KEYS[1], 'mode'); " +
                            "if (mode == false) then " +
                                  "redis.call('hset', KEYS[1], 'mode', 'write'); " +
                                  "redis.call('hset', KEYS[1], ARGV[2], 1); " +
                                  "redis.call('pexpire', KEYS[1], ARGV[1]); " +
                                  "return nil; " +
                              "end; " +
                              "if (mode == 'write') then " +
                                  "if (redis.call('hexists', KEYS[1], ARGV[2]) == 1) then " +
                                      "redis.call('hincrby', KEYS[1], ARGV[2], 1); " + 
                                      "local currentExpire = redis.call('pttl', KEYS[1]); " +
                                      "redis.call('pexpire', KEYS[1], currentExpire + ARGV[1]); " +
                                      "return nil; " +
                                  "end; " +
                                "end;" +
                                "return redis.call('pttl', KEYS[1]);",
                        Arrays.<Object>asList(getName()), 
                        internalLockLeaseTime, getLockName(threadId));
    }
}

还是像上面一样,一行行来分析每句lua脚本执行语义。

客户端A先加读写、再加写锁

KEYS和ARGV参数:

  • KEYS1 = anyLock
  • ARGV1 = 30000
  • ARGV[2] = UUID_01:threadId_01:write
  1. hget anyLock mode,此时没人加锁,mode=false
  2. hset anyLock mode write
  3. hset anyLock UUID_01:threadId_01:write 1
  4. pexpire anyLock 30000

此时redis中数据格式为:

anyLock: {
    "mode": "write",
    "UUID_01:threadId_01:write": 1
}

此时再次来加写锁,直接到另一个if语句中:

  1. hexists anyLock UUID_01:threadId_01:write
  2. hincrby anyLock UUID_01:threadId_01:write 1
  3. pexpire anyLock pttl + 30000

此时redis中数据格式为:

anyLock: {
    "mode": "write",
    "UUID_01:threadId_01:write": 2
}

客户端A和客户端B,先后加读锁,客户端C来加写锁

读锁加完后,此时redis数据格式为:

anyLock: {
  "mode": "read",
  "UUID_01:threadId_01": 1,
  "UUID_02:threadId_02": 1
}

{anyLock}:UUID_01:threadId_01:rwlock_timeout:1    1
{anyLock}:UUID_02:threadId_02:rwlock_timeout:1    1

客户端C参数为:

  • KEYS1 = anyLock
  • ARGV1 = 30000
  • ARGV[2] = UUID_03:threadId_03:write

hget anyLock mode,mode = read,已经有人加了读锁,不是写锁,此时会直接执行:pttl
anyLock,返回一个anyLock的剩余生存时间

  1. hget anyLock mode,mode = read,已经有人加了读锁,不是写锁,所以if语句不会成立
  2. pttl anyLock,返回一个anyLock的剩余生存时间

客户端C加锁失败,就会不断的尝试重试去加锁

客户端A先加写锁、客户端B接着加读锁

加完写锁后此时Redis数据格式为:

anyLock: {
  "mode": "write",
  "UUID_01:threadId_01:write": 1
}

客户端B执行读锁逻辑参数为:

  • KEYS1 = anyLock
  • KEYS[2] = {anyLock}:UUID_02:threadId_02:rwlock_timeout
  • ARGV1 = 30000毫秒
  • ARGV[2] = UUID_02:threadId_02
  • ARGV[3] = UUID_02:threadId_02:write

接着看下加锁逻辑:

image.png

如上图,客户端B加读锁会走到红框中的if逻辑:

  1. hget anyLock mode,mode = write
    客户端A已经加了一个写锁
  2. hexists anyLock UUID_02:threadId_02:write,存在的话,如果客户端B自己之前加过写锁的话,此时才能进入这个分支
  3. 返回pttl anyLock,导致加锁失败

客户端A先加写锁、客户端A接着加读锁

还是接着上面的逻辑,继续分析:

  1. hget anyLock mode,mode = write
    客户端A已经加了一个写锁
  2. hexists anyLock UUID_01:threadId_01:write,此时存在这个key,所以可以进入if分支
  3. hincrby anyLock UUID_01:threadId_01 1,也就是说此时,加了一个读锁
  4. set {anyLock}:UUID_01:threadId_01:rwlock_timeout:1 1,
  5. pexpire anyLock 30000
  6. pexpire {anyLock}:UUID_01:threadId_01:rwlock_timeout:1 30000

此时redis中数据格式为:

anyLock: {
  "mode": "write",
  "UUID_01:threadId_01:write": 1,
  "UUID_01:threadId_01": 1
}

{anyLock}:UUID_01:threadId_01:rwlock_timeout:1    1

客户端A先加读锁、客户端A接着加写锁

客户端A加读锁后,redis中数据结构为:

anyLock: {
  "mode": "read",
  "UUID_01:threadId_01": 1
}

{anyLock}:UUID_01:threadId_01:rwlock_timeout:1  1

此时客户端A再来加写锁,逻辑如下:

image.png

此时客户端A先加的读锁,mode=read,所以再次加写锁是不能成功的

如果是同一个客户端同一个线程,先加了一次写锁,然后加读锁,是可以加成功的,默认是在同一个线程写锁的期间,可以多次加读锁

而同一个客户端同一个线程,先加了一次读锁,是不允许再被加写锁的

总结

显然还有写锁与写锁互斥的逻辑就不分析了,通过上面一些场景的分析,我们可以知道:

  • 读锁与读锁非互斥
  • 读锁与写锁互斥
  • 写锁与写锁互斥
  • 读读、写写 同个客户端同个线程都可重入
  • 先写锁再加读锁可重入
  • 先读锁再写锁不可重入

Redisson读写锁释放原理

Redission 读锁释放原理

不同客户端加了读锁 / 同一个客户端+线程多次可重入加了读锁

例如客户端A先加读锁,然后再次加读锁
最后客户端B来加读锁

此时Redis中数据格式为:

anyLock: {
  "mode": "read",
  "UUID_01:threadId_01": 2,
  "UUID_02:threadId_02": 1
}

{anyLock}:UUID_01:threadId_01:rwlock_timeout:1        1
{anyLock}:UUID_01:threadId_01:rwlock_timeout:2        1
{anyLock}:UUID_02:threadId_02:rwlock_timeout:1        1

接着我们看下释放锁的核心代码:

public class RedissonReadLock extends RedissonLock implements RLock {
    @Override
    protected RFuture<Boolean> unlockInnerAsync(long threadId) {
        String timeoutPrefix = getReadWriteTimeoutNamePrefix(threadId);
        String keyPrefix = getKeyPrefix(threadId, timeoutPrefix);

        return commandExecutor.evalWriteAsync(getName(), LongCodec.INSTANCE, RedisCommands.EVAL_BOOLEAN,
                "local mode = redis.call('hget', KEYS[1], 'mode'); " +
                "if (mode == false) then " +
                    "redis.call('publish', KEYS[2], ARGV[1]); " +
                    "return 1; " +
                "end; " +
                "local lockExists = redis.call('hexists', KEYS[1], ARGV[2]); " +
                "if (lockExists == 0) then " +
                    "return nil;" +
                "end; " +

                "local counter = redis.call('hincrby', KEYS[1], ARGV[2], -1); " + 
                "if (counter == 0) then " +
                    "redis.call('hdel', KEYS[1], ARGV[2]); " + 
                "end;" +
                "redis.call('del', KEYS[3] .. ':' .. (counter+1)); " +

                "if (redis.call('hlen', KEYS[1]) > 1) then " +
                    "local maxRemainTime = -3; " + 
                    "local keys = redis.call('hkeys', KEYS[1]); " + 
                    "for n, key in ipairs(keys) do " + 
                        "counter = tonumber(redis.call('hget', KEYS[1], key)); " + 
                        "if type(counter) == 'number' then " + 
                            "for i=counter, 1, -1 do " + 
                                "local remainTime = redis.call('pttl', KEYS[4] .. ':' .. key .. ':rwlock_timeout:' .. i); " + 
                                "maxRemainTime = math.max(remainTime, maxRemainTime);" + 
                            "end; " + 
                        "end; " + 
                    "end; " +

                    "if maxRemainTime > 0 then " +
                        "redis.call('pexpire', KEYS[1], maxRemainTime); " +
                        "return 0; " +
                    "end;" + 

                    "if mode == 'write' then " + 
                        "return 0;" + 
                    "end; " +
                "end; " +

                "redis.call('del', KEYS[1]); " +
                "redis.call('publish', KEYS[2], ARGV[1]); " +
                "return 1; ",
                Arrays.<Object>asList(getName(), getChannelName(), timeoutPrefix, keyPrefix), 
                LockPubSub.unlockMessage, getLockName(threadId));
    }
}

客户端A来释放锁:
对应的KEYS和ARGV参数为:

  • KEYS1 = anyLock

  • KEYS[2] = redisson_rwlock:{anyLock}

  • KEYS[3] = {anyLock}:UUID_01:threadId_01:rwlock_timeout

  • KEYS[4] = {anyLock}

  • ARGV1 = 0

  • ARGV[2] = UUID_01:threadId_01

接下来开始执行操作:

  1. hget anyLock mode,mode = read
  2. hexists anyLock UUID_01:threadId_01,肯定是存在的,因为这个客户端A加过读锁
  3. hincrby anyLock UUID_01:threadId_01 -1,将这个客户端对应的加锁次数递减1,现在就是变成1,counter = 1
  4. del {anyLock}:UUID_01:threadId_01:rwlock_timeout:2,删除了一个timeout key

此时Redis中的数据结构为:

anyLock: {
  "mode": "read",
  "UUID_01:threadId_01": 1,
  "UUID_02:threadId_02": 1
}

{anyLock}:UUID_01:threadId_01:rwlock_timeout:1    1
{anyLock}:UUID_02:threadId_02:rwlock_timeout:1    1

此时继续往下,具体逻辑如图:

image.png

  1. hlen anyLock > 1,就是hash里面的元素超过1个
  2. pttl {anyLock}:UUID_01:threadId_01:rwlock_timeout:1,此时获取那个timeout key的剩余生存时间还有多少毫秒,比如说此时这个key的剩余生存时间是20000毫秒

这个for循环的含义是获取到了所有的timeout key的最大的一个剩余生存时间,假设最大的剩余生存时间是25000毫秒

客户端A继续来释放锁:

此时客户端A执行流程还会和上面一直,执行完成后Redis中数据结构为:

anyLock: {
  "mode": "read",
  "UUID_02:threadId_02": 1
}

{anyLock}:UUID_02:threadId_02:rwlock_timeout:1    1

因为这里会走counter == 0的逻辑,所以会执行"redis.call('hdel', KEYS[1], ARGV[2]); "

客户端B继续来释放锁:

客户端B流程也和上面一直,执行完后就会删除anyLock这个key

同一个客户端/线程先加写锁再加读锁

上面已经分析过这种情形,操作过后Redis中数据结构为:

anyLock: {
  "mode": "write",
  "UUID_01:threadId_01:write": 1,
  "UUID_01:threadId_01": 1
}

{anyLock}:UUID_01:threadId_01:rwlock_timeout:1    1

此时客户端A来释放读锁:

  1. hincrby anyLock UUID_01:threadId_01 -1,将这个客户端对应的加锁次数递减1,现在就是变成1,counter = 0
  2. hdel anyLock UUID_01:threadId_01,此时就是从hash数据结构中删除客户端A这个加锁的记录
  3. del {anyLock}:UUID_01:threadId_01:rwlock_timeout:1,删除了一个timeout key

此时Redis中数据变成:

anyLock: {
  "mode": "write",
  "UUID_01:threadId_01:write": 1
}

Redisson写锁释放原理

先看下写锁释放的核心逻辑:

public class RedissonWriteLock extends RedissonLock implements RLock {
    @Override
    protected RFuture<Boolean> unlockInnerAsync(long threadId) {
        return commandExecutor.evalWriteAsync(getName(), LongCodec.INSTANCE, RedisCommands.EVAL_BOOLEAN,
                "local mode = redis.call('hget', KEYS[1], 'mode'); " +
                "if (mode == false) then " +
                    "redis.call('publish', KEYS[2], ARGV[1]); " +
                    "return 1; " +
                "end;" +
                "if (mode == 'write') then " +
                    "local lockExists = redis.call('hexists', KEYS[1], ARGV[3]); " +
                    "if (lockExists == 0) then " +
                        "return nil;" +
                    "else " +
                        "local counter = redis.call('hincrby', KEYS[1], ARGV[3], -1); " +
                        "if (counter > 0) then " +
                            "redis.call('pexpire', KEYS[1], ARGV[2]); " +
                            "return 0; " +
                        "else " +
                            "redis.call('hdel', KEYS[1], ARGV[3]); " +
                            "if (redis.call('hlen', KEYS[1]) == 1) then " +
                                "redis.call('del', KEYS[1]); " +
                                "redis.call('publish', KEYS[2], ARGV[1]); " + 
                            "else " +
                                // has unlocked read-locks
                                "redis.call('hset', KEYS[1], 'mode', 'read'); " +
                            "end; " +
                            "return 1; "+
                        "end; " +
                    "end; " +
                "end; "
                + "return nil;",
        Arrays.<Object>asList(getName(), getChannelName()), 
        LockPubSub.unlockMessage, internalLockLeaseTime, getLockName(threadId));
    }
}
同一个客户端多次可重入加写锁 / 同一个客户端先加写锁再加读锁

客户端A加两次写锁释放:

此时Redis中数据为:

anyLock: {
  "mode": "write",
  "UUID_01:threadId_01:write": 2,
  "UUID_01:threadId_01": 1
}

{anyLock}:UUID_01:threadId_01:rwlock_timeout:1    1

客户端A来释放锁KEYS和ARGV参数:

  • KEYS1 = anyLock

  • KEYS[2] = redisson_rwlock:{anyLock}

  • ARGV1 = 0

  • ARGV[2] = 30000

  • ARGV[3] = UUID_01:threadId_01:write

直接分析lua代码:

  1. 上面mode=write,后面使用hincrby进行-1操作,此时count=1
  2. 如果count>0,此时使用pexpire然后返回0
  3. 此时客户端A再来释放写锁,count=0
  4. hdel anyLock UUID_01:threadId_01:write

此时Redis中数据:

anyLock: {
  "mode": "write",
  "UUID_01:threadId_01": 1
}

{anyLock}:UUID_01:threadId_01:rwlock_timeout:1    1

后续还会接着判断,如果count=0,代表写锁都已经释放完了,此时hlen如果>1,代表加的还有读锁,所以接着执行:hset anyLock mode read, 将写锁转换为读锁

最终Redis数据为:

anyLock: {
  "mode": "read",
  "UUID_01:threadId_01": 1
}

{anyLock}:UUID_01:threadId_01:rwlock_timeout:1    1

总结

Redisson陆续也更新了好几篇了,疫情期间宅在家里一直学习Redisson相关内容,这篇文章写了2天,从早到晚。

读写锁这块内容真的很多,本篇篇幅很长,如果学习本篇文章最好跟着源码一起读,后续还会继续更新Redisson相关内容,如有不正确的地方,欢迎指正!

申明

本文章首发自本人博客:https://www.cnblogs.com/wang-meng 和公众号:壹枝花算不算浪漫,如若转载请标明来源!

感兴趣的小伙伴可关注个人公众号:壹枝花算不算浪漫

加载全部内容

相关教程
猜你喜欢
用户评论