什么是分布式锁
当多个进程在同一个系统中,用分布式锁控制多个进程对资源的访问
分布式锁应用场景- 传统的单体应用单机部署情况下,可以使用java并发处理相关的API进行互斥控制。
- 分布式系统后由于多线程,多进程分布在不同机器上,使单机部署情况下的并发控制锁策略失效,为了解决跨JVM互斥机制来控制共享资源的访问,这就是分布式锁的来源;分布式锁应用场景大都是高并发、大流量场景。
- 加锁机制:根据hash节点选择一个客户端执行lua脚本
- 锁互斥机制:再来一个客户端执行同样的lua脚本会提示已经存在锁,然后进入循环一直尝试加锁
- 可重入机制
- watch dog自动延期机制
- 释放锁机制
测试用例
单机
1 private RedissonClient getClient(){ 2 Config config = new Config(); 3 config.useSingleServer().setAddress("redis://127.0.0.1:6379");//.setPassword("");//.setConnectionMinimumIdleSize(10).setConnectionPoolSize(10);//.setConnectionPoolSize();//172.16.10.164 4 RedissonClient redissonClient = Redisson.create(config); 5 return redissonClient; 6 } 7 private ExecutorService executorService = Executors.newCachedThreadPool(); 8 ---------------------------------------------------------------- 9 int[] count = {0}; 10 for (int i = 0; i < 10; i++) { 11 RedissonClient client = getClient(); 12 final RedisLock redisLock = new RedisLock(client,"lock_key"); 13 executorService.submit(() -> { 14 try { 15 redisLock.lock(); 16 count[0]++; 17 } catch (Exception e) { 18 e.printStackTrace(); 19 } finally { 20 try { 21 redisLock.unlock(); 22 } catch (Exception e) { 23 e.printStackTrace(); 24 } 25 } 26 });
RedLock
1 public static RLock create (String url, String key){ 2 Config config = new Config(); 3 config.useSingleServer().setAddress(url); 4 RedissonClient redissonClient = Redisson.create(config); 5 return redissonClient.getLock(key); 6 } 7 8 RedissonRedLock redissonRedLock = new RedissonRedLock( 9 create("redis://redis://127.0.0.1:6379","lock_key1"), 10 create("redis://redis://127.0.0.1:6380","lock_key2"), 11 create("redis://redis://127.0.0.1:6381","lock_key3")); 12 RedisRedLock redLock = new RedisRedLock(redissonRedLock); 13 14 private ExecutorService executorService = Executors.newCachedThreadPool(); 15 16 ------------------------------------------------------------------ 17 int[] count = {0}; 18 for (int i = 0; i < 2; i++) { 19 executorService.submit(() -> { 20 try { 21 redLock.lock(); 22 count[0]++; 23 } catch (Exception e) { 24 e.printStackTrace(); 25 } finally { 26 try { 27 redLock.unlock(); 28 } catch (Exception e) { 29 e.printStackTrace(); 30 } 31 } 32 }); 33 }
redis源码分析
redisson
1 public void lockInterruptibly(long leaseTime, TimeUnit unit) throws InterruptedException { 2 //获取当前线程id 3 long threadId = Thread.currentThread().getId(); 4 //尝试获得锁,返回还剩余的锁过期时间 5 Long ttl = tryAcquire(leaseTime, unit, threadId); 6 // lock acquired 7 //如果ttl为空,代表当前没有锁,获取成功 8 if (ttl == null) { 9 return; 10 } 11 12 //如果获取锁失败,则订阅到对应这个锁的channel,一旦其他线程释放锁时,通知线程去获取锁 13 RFuture<RedissonLockEntry> future = subscribe(threadId); 14 commandExecutor.syncSubscription(future); 15 16 try { 17 //循环等待 18 while (true) { 19 //尝试获得锁 20 ttl = tryAcquire(leaseTime, unit, threadId); 21 // lock acquired 22 if (ttl == null) { 23 break; 24 } 25 26 // waiting for message //ttl大于0,则等待ttl时间后继续尝试获取锁 27 if (ttl >= 0) { 28 getEntry(threadId).getLatch().tryAcquire(ttl, TimeUnit.MILLISECONDS); 29 } else { 30 getEntry(threadId).getLatch().acquire(); 31 } 32 } 33 } finally { 34 unsubscribe(future, threadId); 35 } 36 // get(lockAsync(leaseTime, unit)); 37 }
2、基于ETCD实现分布式锁分析 ETCD分布式锁的实现
-
Lease机制:租约机制(TTL,Time To Live),Etcd 可以为存储的 key-value 对设置租约,
当租约到期,key-value 将失效删除;同时也支持续约,通过客户端可以在租约到期之前续约,
以避免 key-value 对过期失效。Lease 机制可以保证分布式锁的安全性,为锁对应的 key 配置租约,
即使锁的持有者因故障而不能主动释放锁,锁也会因租约到期而自动释放 -
Revision机制:每个 key 带有一个 Revision 号,每进行一次事务加一,它是全局唯一的,
通过 Revision 的大小就可以知道进行写操作的顺序。在实现分布式锁时,多个客户端同时抢锁,
根据 Revision 号大小依次获得锁,可以避免 “羊群效应” ,实现公平锁 -
Prefix机制:即前缀机制。例如,一个名为 /etcdlock 的锁,两个争抢它的客户端进行写操作,
实际写入的 key 分别为:key1="/etcdlock/UUID1",key2="/etcdlock/UUID2",
其中,UUID 表示全局唯一的 ID,确保两个 key 的唯一性。写操作都会成功,但返回的 Revision 不一样,
那么,如何判断谁获得了锁呢?通过前缀 /etcdlock 查询,返回包含两个 key-value 对的的 KeyValue 列表,
同时也包含它们的 Revision,通过 Revision 大小,客户端可以判断自己是否获得锁 -
Watch机制:即监听机制,Watch 机制支持 Watch 某个固定的 key,也支持 Watch 一个范围(前缀机制),
当被 Watch 的 key 或范围发生变化,客户端将收到通知;在实现分布式锁时,如果抢锁失败,
可通过 Prefix 机制返回的 KeyValue 列表获得 Revision 比自己小且相差最小的 key(称为 pre-key),
对 pre-key 进行监听,因为只有它释放锁,自己才能获得锁,如果 Watch 到 pre-key 的 DELETE 事件,
则说明 pre-key 已经释放,自己已经持有锁
客户端连接 Etcd,以 /etcd/lock 为前缀创建全局唯一的 key,
假设第一个客户端对应的 key="/etcd/lock/UUID1",第二个为 key="/etcd/lock/UUID2";
客户端分别为自己的 key 创建租约 - Lease,租约的长度根据业务耗时确定;
当一个客户端持有锁期间,其它客户端只能等待,为了避免等待期间租约失效,
客户端需创建一个定时任务作为“心跳”进行续约。此外,如果持有锁期间客户端崩溃,
心跳停止,key 将因租约到期而被删除,从而锁释放,避免死锁
执行 put 操作,将步骤 1 中创建的 key 绑定租约写入 Etcd,根据 Etcd 的 Revision 机制,
假设两个客户端 put 操作返回的 Revision 分别为 1、2,客户端需记录 Revision 用以
接下来判断自己是否获得锁
客户端以前缀 /etcd/lock/ 读取 keyValue 列表,判断自己 key 的 Revision 是否为当前列表中
最小的,如果是则认为获得锁;否则监听列表中前一个 Revision 比自己小的 key 的删除事件,一旦监听到删除事件或者因租约失效而删除的事件,则自己获得锁。
获得锁后,操作共享资源,执行业务代码
步骤 6、释放锁完成业务流程后,删除对应的key释放锁
测试用例1 public class EtcdDistributeLock extends AbstractLock{ 2 3 private Client client; 4 private Lock lockClient; 5 private Lease leaseClient; 6 private String lockKey; 7 private String lockPath; 8 /** 锁的次数 */ 9 private AtomicInteger lockCount; 10 /** 租约有效期,防止客户端崩溃,可在租约到期后自动释放锁;另一方面,正常执行过程中,会自动进行续租,单位 ns */ 11 private Long leaseTTL; 12 /** 续约锁租期的定时任务,初次启动延迟,单位默认为 s,默认为1s,可根据业务定制设置*/ 13 private Long initialDelay = 0L; 14 /** 定时任务线程池类 */ 15 ScheduledExecutorService service = null; 16 /** 保存线程与锁对象的映射,锁对象包含重入次数,重入次数的最大限制为Int的最大值 */ 17 private final ConcurrentMap<Thread, LockData> threadData = Maps.newConcurrentMap(); 18 19 public EtcdDistributeLock(){} 20 21 public EtcdDistributeLock(Client client, String lockKey, long leaseTTL,TimeUnit unit){ 22 this.client = client; 23 lockClient = client.getLockClient(); 24 leaseClient = client.getLeaseClient(); 25 this.lockKey = lockKey; 26 // 转纳秒 27 this.leaseTTL = unit.toNanos(leaseTTL); 28 service = Executors.newSingleThreadScheduledExecutor(); 29 } 30 31 32 @Override 33 public void lock() { 34 // 检查重入性 35 Thread currentThread = Thread.currentThread(); 36 LockData oldLockData = threadData.get(currentThread); 37 if (oldLockData != null && oldLockData.isLockSuccess()) { 38 // re-entering 39 int lockCount = oldLockData.lockCount.incrementAndGet(); 40 if(lockCount < 0 ){ 41 throw new Error("超出可重入次数限制"); 42 } 43 return; 44 } 45 46 // 记录租约 ID 47 Long leaseId = 0L; 48 try{ 49 leaseId = leaseClient.grant(TimeUnit.NANOSECONDS.toSeconds(leaseTTL)).get().getID(); 50 // 续租心跳周期 51 long period = leaseTTL - leaseTTL / 5; 52 // 启动定时任务续约 53 service.scheduleAtFixedRate(new EtcdDistributeLock.KeepAliveRunnable(leaseClient, leaseId), 54 initialDelay,period,TimeUnit.NANOSECONDS); 55 LockResponse lockResponse = lockClient.lock(ByteSequence.from(lockKey.getBytes()), leaseId).get(); 56 if(lockResponse != null){ 57 lockPath = lockResponse.getKey().toString(Charset.forName("utf-8")); 58 log.info("获取锁成功,锁路径:{},线程:{}",lockPath,currentThread.getName()); 59 } 60 }catch (InterruptedException | ExecutionException e){ 61 log.error("获取锁失败",e); 62 return; 63 } 64 // 获取锁成功,锁对象设置 65 LockData newLockData = new LockData(currentThread, lockKey); 66 newLockData.setLeaseId(leaseId); 67 newLockData.setService(service); 68 threadData.put(currentThread, newLockData); 69 newLockData.setLockSuccess(true); 70 } 71 72 @Override 73 public void lockInterruptibly() throws InterruptedException { 74 super.lockInterruptibly(); 75 } 76 77 @Override 78 public boolean tryLock() { 79 return super.tryLock(); 80 } 81 82 @Override 83 public boolean tryLock(long time, TimeUnit unit) throws InterruptedException { 84 return super.tryLock(time,unit); 85 } 86 87 88 @Override 89 public void unlock() { 90 Thread currentThread = Thread.currentThread(); 91 LockData lockData = threadData.get(currentThread); 92 if (lockData == null){ 93 throw new IllegalMonitorStateException("You do not own the lock: " + lockKey); 94 } 95 int newLockCount = lockData.lockCount.decrementAndGet(); 96 if ( newLockCount > 0 ) { 97 return; 98 } 99 if ( newLockCount < 0 ) { 100 throw new IllegalMonitorStateException("Lock count has gone negative for lock: " + lockKey); 101 } 102 try { 103 // 释放锁 104 if(lockPath != null){ 105 lockClient.unlock(ByteSequence.from(lockPath.getBytes())).get(); 106 } 107 if(lockData != null){ 108 // 关闭定时任务 109 lockData.getService().shutdown(); 110 // 删除租约 111 if (lockData.getLeaseId() != 0L) { 112 leaseClient.revoke(lockData.getLeaseId()); 113 } 114 } 115 } catch (InterruptedException | ExecutionException e) { 116 log.error("解锁失败",e); 117 }finally { 118 // 移除当前线程资源 119 threadData.remove(currentThread); 120 } 121 } 122 123 124 @Override 125 public Condition newCondition() { 126 return super.newCondition(); 127 } 128 129 /** 130 * 心跳续约线程类 131 */ 132 public static class KeepAliveRunnable implements Runnable { 133 private Lease leaseClient; 134 private long leaseId; 135 136 public KeepAliveRunnable(Lease leaseClient, long leaseId) { 137 this.leaseClient = leaseClient; 138 this.leaseId = leaseId; 139 } 140 141 @Override 142 public void run() { 143 // 对该leaseid进行一次续约 144 leaseClient.keepAliveOnce(leaseId); 145 } 146 } 147 148 public class EtcdLockTest { 149 private Client client; 150 private String key = "/etcd/lock"; 151 private static final String server = "http://xxxx:xxxx"; 152 private ExecutorService executorService = Executors.newFixedThreadPool(10000); 153 154 @Before 155 public void before() throws Exception { 156 initEtcdClient(); 157 } 158 159 private void initEtcdClient(){ 160 client = Client.builder().endpoints(server).build(); 161 } 162 163 @Test 164 public void testEtcdDistributeLock() throws InterruptedException { 165 int[] count = {0}; 166 for (int i = 0; i < 100; i++) { 167 executorService.submit(() -> { 168 final EtcdDistributeLock lock = new EtcdDistributeLock(client, key,20,TimeUnit.SECONDS); 169 try { 170 lock.lock(); 171 count[0]++; 172 } catch (Exception e) { 173 e.printStackTrace(); 174 } finally { 175 try { 176 lock.unlock(); 177 } catch (Exception e) { 178 e.printStackTrace(); 179 } 180 } 181 }); 182 } 183 executorService.shutdown(); 184 executorService.awaitTermination(1, TimeUnit.HOURS); 185 System.err.println("执行结果: " + count[0]); 186 } 187 }
3、基于Zookeeper分布式锁 实现原理
- 启动客户端,确认链接到了服务器
- 多个客户端并发的在特定路径下创建临时性顺序节点
- 客户端判断自己的创建的顺序节点是否是最小的,如果是最小的,则获取锁成功
- 第三步若判定失败,则采用zk的watch机制监听自己的前一个顺序节点,等待前一个节点的删除(放锁)事件,再开始第三步判定。
zookeeper作为高性能分布式协调框架,可以把其看做一个文件系统,其中有节点的概念,并且分为4种:1.持久性节点2.持久性顺序节点3.临时性节点4.临时性顺序节点。
分布式锁的实现主要思路就是:监控其他客户端的状态,来判断自己是否可以获得锁。
采用临时性顺序节点的原因:
- zk服务器维护了客户端的会话有效性,当会话失效的时候,其会话所创建的临时性节点都会被删除,通过这一特点,可以通过watch临时节点来监控其他客户端的情况,方便自己做出相应动作。
- 因为zk对写操作是顺序性的,所以并发创建的顺序节点会有一个唯一确定的序号,当前锁是公平锁的一种实现,所以依靠这种顺序性可以很好的解释—节点序列小的获取到锁并且可以采用watch自己的前一个节点来避免惊群现象(这样watch事件的传播是线性的)。
1 public class ZKLock extends AbstractLock { 2 3 /** 4 * 1.Connect to zk 5 */ 6 private CuratorFramework client; 7 8 private InterProcessLock lock ; 9 10 11 public ZKLock(String zkAddress,String lockPath) { 12 // 1.Connect to zk 13 client = CuratorFrameworkFactory.newClient( 14 zkAddress, 15 new RetryNTimes(5, 5000) 16 ); 17 client.start(); 18 if(client.getState() == CuratorFrameworkState.STARTED){ 19 log.info("zk client start successfully!"); 20 log.info("zkAddress:{},lockPath:{}",zkAddress,lockPath); 21 }else{ 22 throw new RuntimeException("客户端启动失败。。。"); 23 } 24 this.lock = defaultLock(lockPath); 25 } 26 27 private InterProcessLock defaultLock(String lockPath ){ 28 return new InterProcessMutex(client, lockPath); 29 } 30 @Override 31 public void lock() { 32 try { 33 this.lock.acquire(); 34 } catch (Exception e) { 35 throw new RuntimeException(e); 36 } 37 } 38 39 @Override 40 public boolean tryLock() { 41 boolean flag ; 42 try { 43 flag=this.lock.acquire(0,TimeUnit.SECONDS); 44 } catch (Exception e) { 45 throw new RuntimeException(e); 46 } 47 return flag; 48 } 49 50 @Override 51 public boolean tryLock(long time, TimeUnit unit) throws InterruptedException { 52 boolean flag ; 53 try { 54 flag=this.lock.acquire(time,unit); 55 } catch (Exception e) { 56 throw new RuntimeException(e); 57 } 58 return flag; 59 } 60 61 @Override 62 public void unlock() { 63 try { 64 this.lock.release(); 65 } catch (Exception e) { 66 throw new RuntimeException(e); 67 } 68 } 69 70 } 71 private ExecutorService executorService = Executors.newCachedThreadPool(); 72 73 74 @Test 75 public void testLock() throws Exception{ 76 ZKLock zkLock = new ZKLock("xxxx:xxxx","/lockPath"); 77 int[] num = {0}; 78 long start = System.currentTimeMillis(); 79 for(int i=0;i<200;i++){ 80 executorService.submit(()->{ 81 try { 82 zkLock.lock(); 83 num[0]++; 84 } catch (Exception e){ 85 throw new RuntimeException(e); 86 } finally { 87 zkLock.unlock(); 88 } 89 }); 90 91 } 92 executorService.shutdown(); 93 executorService.awaitTermination(1, TimeUnit.HOURS); 94 log.info("耗时:{}",System.currentTimeMillis()-start); 95 System.out.println(num[0]); 96 }