Spring Integration Redis 分布式锁实现详解

在分布式微服务体系大行其道的今天，分布式锁的概念已经深入每个程序猿的内心，实现分布式锁的方式也有很多种，例如使用数据库、Redis、zookeeper都可以实现分布式锁，今天我们来分析一下使用Redis来实现分布式锁的底层实现原理。

在分布式场景中，存在这么一种可能：多个实例需要互斥访问共享资源，最典型的案例就是秒杀，瞬时流量高、标的数量有限，要防止超卖的情况发生。

分布式锁一般具有以下特征：

• 互斥性：同一时刻只能有一个线程持有锁
• 可重入性：同一节点上的同一个线程获取到锁之后，还能再次获取锁
• 锁超时：持有锁的线程不能始终持有锁，必须要有超时机制，防止死锁
• 高性能和高可用性：加锁和解锁需要高效，不能占用太多系统资源和处理时间，高可用需要防止锁失效情况的发生
• 响应中断：可以及时从阻塞状态中唤醒

##Redis对分布式锁实现的理论底层支持

关于利用Redis来实现分布式锁理论知识，可以参考小米信息部技术团队的文章，这里就不赘述了链接如下： xiaomi-info.github.io/2019/12/17/…

我们今天主要讲的是如何使用Spring Integration 这个项目来实现Redis分布式锁。

Spring Integration项目支持基于Spring的应用程序中的轻量级消息传递，并通过声明性适配器支持与外部系统的集成。这些适配器在Spring对远程、消息传递和调度的支持之上提供了更高级别的抽象。Spring Integration的主要目标是为构建企业集成解决方案提供一个简单的模型，同时维护对生成可维护、可测试代码至关重要的关注点分离。 RedisLockRegistry 类源码：

package org.springframework.integration.redis.util;

import java.text.SimpleDateFormat;
import java.util.Collections;
import java.util.Date;
import java.util.Map;
import java.util.UUID;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.Executor;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.locks.Condition;
import java.util.concurrent.locks.Lock;
import java.util.concurrent.locks.ReentrantLock;

import org.apache.commons.logging.Log;
import org.apache.commons.logging.LogFactory;

import org.springframework.beans.factory.DisposableBean;
import org.springframework.dao.CannotAcquireLockException;
import org.springframework.data.redis.connection.RedisConnectionFactory;
import org.springframework.data.redis.core.StringRedisTemplate;
import org.springframework.data.redis.core.script.DefaultRedisScript;
import org.springframework.data.redis.core.script.RedisScript;
import org.springframework.integration.support.locks.ExpirableLockRegistry;
import org.springframework.scheduling.concurrent.CustomizableThreadFactory;
import org.springframework.util.Assert;
import org.springframework.util.ReflectionUtils;

/**
 * Implementation of {@link ExpirableLockRegistry} providing a distributed lock using Redis.
 * Locks are stored under the key {@code registryKey:lockKey}. Locks expire after
 * (default 60) seconds. Threads unlocking an
 * expired lock will get an {@link IllegalStateException}. This should be
 * considered as a critical error because it is possible the protected
 * resources were compromised.
 * <p>
 * Locks are reentrant.
 * <p>
 * <b>However, locks are scoped by the registry; a lock from a different registry with the
 * same key (even if the registry uses the same 'registryKey') are different
 * locks, and the second cannot be acquired by the same thread while the first is
 * locked.</b>
 * <p>
 * <b>Note: This is not intended for low latency applications.</b> It is intended
 * for resource locking across multiple JVMs.
 * <p>
 * {@link Condition}s are not supported.
 *
 * @author Gary Russell
 * @author Konstantin Yakimov
 * @author Artem Bilan
 * @author Vedran Pavic
 *
 * @since 4.0
 *
 */
public final class RedisLockRegistry implements ExpirableLockRegistry, DisposableBean {

	private static final Log LOGGER = LogFactory.getLog(RedisLockRegistry.class);

	private static final long DEFAULT_EXPIRE_AFTER = 60000L;

	private static final String OBTAIN_LOCK_SCRIPT =
			"local lockClientId = redis.call('GET', KEYS[1])\n" +
					"if lockClientId == ARGV[1] then\n" +
					"  redis.call('PEXPIRE', KEYS[1], ARGV[2])\n" +
					"  return true\n" +
					"elseif not lockClientId then\n" +
					"  redis.call('SET', KEYS[1], ARGV[1], 'PX', ARGV[2])\n" +
					"  return true\n" +
					"end\n" +
					"return false";


	private final Map<String, RedisLock> locks = new ConcurrentHashMap<>();

	private final String clientId = UUID.randomUUID().toString();

	private final String registryKey;

	private final StringRedisTemplate redisTemplate;

	private final RedisScript<Boolean> obtainLockScript;

	private final long expireAfter;

	/**
	 * An {@link ExecutorService} to call {@link StringRedisTemplate#delete} in
	 * the separate thread when the current one is interrupted.
	 */
	private Executor executor =
			Executors.newCachedThreadPool(new CustomizableThreadFactory("redis-lock-registry-"));

	/**
	 * Flag to denote whether the {@link ExecutorService} was provided via the setter and
	 * thus should not be shutdown when {@link #destroy()} is called
	 */
	private boolean executorExplicitlySet;

	private volatile boolean unlinkAvailable = true;

	/**
	 * Constructs a lock registry with the default (60 second) lock expiration.
	 * @param connectionFactory The connection factory.
	 * @param registryKey The key prefix for locks.
	 */
	public RedisLockRegistry(RedisConnectionFactory connectionFactory, String registryKey) {
		this(connectionFactory, registryKey, DEFAULT_EXPIRE_AFTER);
	}

	/**
	 * Constructs a lock registry with the supplied lock expiration.
	 * @param connectionFactory The connection factory.
	 * @param registryKey The key prefix for locks.
	 * @param expireAfter The expiration in milliseconds.
	 */
	public RedisLockRegistry(RedisConnectionFactory connectionFactory, String registryKey, long expireAfter) {
		Assert.notNull(connectionFactory, "'connectionFactory' cannot be null");
		Assert.notNull(registryKey, "'registryKey' cannot be null");
		this.redisTemplate = new StringRedisTemplate(connectionFactory);
		this.obtainLockScript = new DefaultRedisScript<>(OBTAIN_LOCK_SCRIPT, Boolean.class);
		this.registryKey = registryKey;
		this.expireAfter = expireAfter;
	}

	/**
	 * Set the {@link Executor}, where is not provided then a default of
	 * cached thread pool Executor will be used.
	 * @param executor the executor service
	 * @since 5.0.5
	 */
	public void setExecutor(Executor executor) {
		this.executor = executor;
		this.executorExplicitlySet = true;
	}

	@Override
	public Lock obtain(Object lockKey) {
		Assert.isInstanceOf(String.class, lockKey);
		String path = (String) lockKey;
		return this.locks.computeIfAbsent(path, RedisLock::new);
	}

	@Override
	public void expireUnusedOlderThan(long age) {
		long now = System.currentTimeMillis();
		this.locks.entrySet()
				.removeIf((entry) -> {
					RedisLock lock = entry.getValue();
					return now - lock.getLockedAt() > age && !lock.isAcquiredInThisProcess();
				});
	}

	@Override
	public void destroy() {
		if (!this.executorExplicitlySet) {
			((ExecutorService) this.executor).shutdown();
		}
	}

	private final class RedisLock implements Lock {

		private final String lockKey;

		private final ReentrantLock localLock = new ReentrantLock();

		private volatile long lockedAt;

		private RedisLock(String path) {
			this.lockKey = constructLockKey(path);
		}

		private String constructLockKey(String path) {
			return RedisLockRegistry.this.registryKey + ':' + path;
		}

		public long getLockedAt() {
			return this.lockedAt;
		}

		@Override
		public void lock() {
			this.localLock.lock();
			while (true) {
				try {
					while (!obtainLock()) {
						Thread.sleep(100); //NOSONAR
					}
					break;
				}
				catch (InterruptedException e) {
					/*
					 * This method must be uninterruptible so catch and ignore
					 * interrupts and only break out of the while loop when
					 * we get the lock.
					 */
				}
				catch (Exception e) {
					this.localLock.unlock();
					rethrowAsLockException(e);
				}
			}
		}

		private void rethrowAsLockException(Exception e) {
			throw new CannotAcquireLockException("Failed to lock mutex at " + this.lockKey, e);
		}

		@Override
		public void lockInterruptibly() throws InterruptedException {
			this.localLock.lockInterruptibly();
			try {
				while (!obtainLock()) {
					Thread.sleep(100); //NOSONAR
				}
			}
			catch (InterruptedException ie) {
				this.localLock.unlock();
				Thread.currentThread().interrupt();
				throw ie;
			}
			catch (Exception e) {
				this.localLock.unlock();
				rethrowAsLockException(e);
			}
		}

		@Override
		public boolean tryLock() {
			try {
				return tryLock(0, TimeUnit.MILLISECONDS);
			}
			catch (InterruptedException e) {
				Thread.currentThread().interrupt();
				return false;
			}
		}

		@Override
		public boolean tryLock(long time, TimeUnit unit) throws InterruptedException {
			long now = System.currentTimeMillis();
			if (!this.localLock.tryLock(time, unit)) {
				return false;
			}
			try {
				long expire = now + TimeUnit.MILLISECONDS.convert(time, unit);
				boolean acquired;
				while (!(acquired = obtainLock()) && System.currentTimeMillis() < expire) { //NOSONAR
					Thread.sleep(100); //NOSONAR
				}
				if (!acquired) {
					this.localLock.unlock();
				}
				return acquired;
			}
			catch (Exception e) {
				this.localLock.unlock();
				rethrowAsLockException(e);
			}
			return false;
		}

		private boolean obtainLock() {
			Boolean success =
					RedisLockRegistry.this.redisTemplate.execute(RedisLockRegistry.this.obtainLockScript,
							Collections.singletonList(this.lockKey), RedisLockRegistry.this.clientId,
							String.valueOf(RedisLockRegistry.this.expireAfter));

			boolean result = Boolean.TRUE.equals(success);

			if (result) {
				this.lockedAt = System.currentTimeMillis();
			}
			return result;
		}

		@Override
		public void unlock() {
			if (!this.localLock.isHeldByCurrentThread()) {
				throw new IllegalStateException("You do not own lock at " + this.lockKey);
			}
			if (this.localLock.getHoldCount() > 1) {
				this.localLock.unlock();
				return;
			}
			try {
				if (!isAcquiredInThisProcess()) {
					throw new IllegalStateException("Lock was released in the store due to expiration. " +
							"The integrity of data protected by this lock may have been compromised.");
				}

				if (Thread.currentThread().isInterrupted()) {
					RedisLockRegistry.this.executor.execute(this::removeLockKey);
				}
				else {
					removeLockKey();
				}

				if (LOGGER.isDebugEnabled()) {
					LOGGER.debug("Released lock; " + this);
				}
			}
			catch (Exception e) {
				ReflectionUtils.rethrowRuntimeException(e);
			}
			finally {
				this.localLock.unlock();
			}
		}

		private void removeLockKey() {
			if (RedisLockRegistry.this.unlinkAvailable) {
				try {
					RedisLockRegistry.this.redisTemplate.unlink(this.lockKey);
				}
				catch (Exception ex) {
					RedisLockRegistry.this.unlinkAvailable = false;
					if (LOGGER.isDebugEnabled()) {
						LOGGER.debug("The UNLINK command has failed (not supported on the Redis server?); " +
								"falling back to the regular DELETE command", ex);
					}
					else {
						LOGGER.warn("The UNLINK command has failed (not supported on the Redis server?); " +
								"falling back to the regular DELETE command: " + ex.getMessage());
					}
					RedisLockRegistry.this.redisTemplate.delete(this.lockKey);
				}
			}
			else {
				RedisLockRegistry.this.redisTemplate.delete(this.lockKey);
			}
		}

		@Override
		public Condition newCondition() {
			throw new UnsupportedOperationException("Conditions are not supported");
		}

		public boolean isAcquiredInThisProcess() {
			return RedisLockRegistry.this.clientId.equals(
					RedisLockRegistry.this.redisTemplate.boundValueOps(this.lockKey).get());
		}

		@Override
		public String toString() {
			SimpleDateFormat dateFormat = new SimpleDateFormat("yyyy-MM-dd@HH:mm:ss.SSS");
			return "RedisLock [lockKey=" + this.lockKey
					+ ",lockedAt=" + dateFormat.format(new Date(this.lockedAt))
					+ ", clientId=" + RedisLockRegistry.this.clientId
					+ "]";
		}

		@Override
		public int hashCode() {
			final int prime = 31;
			int result = 1;
			result = prime * result + getOuterType().hashCode();
			result = prime * result + ((this.lockKey == null) ? 0 : this.lockKey.hashCode());
			result = prime * result + (int) (this.lockedAt ^ (this.lockedAt >>> 32));
			result = prime * result + RedisLockRegistry.this.clientId.hashCode();
			return result;
		}

		@Override
		public boolean equals(Object obj) {
			if (this == obj) {
				return true;
			}
			if (obj == null) {
				return false;
			}
			if (getClass() != obj.getClass()) {
				return false;
			}
			RedisLock other = (RedisLock) obj;
			if (!getOuterType().equals(other.getOuterType())) {
				return false;
			}
			if (!this.lockKey.equals(other.lockKey)) {
				return false;
			}
			return this.lockedAt == other.lockedAt;
		}

		private RedisLockRegistry getOuterType() {
			return RedisLockRegistry.this;
		}

	}

}

首先我们看一下该类的javadoc,通过阅读javadoc,可以快速了解作者的意图，与作者实现灵魂上的沟通，我们自己在写代码时，也应该将设计意图使用javadoc的方式来体现，让后继的维护者可以了解当时的设计意图，也方便自己在未来需要重新修改代码时，快速回忆自己的设计方案。

RedisLockRegsitry 实现了ExpirableLockRegistry 接口，基于Redis实现了分布式锁，锁存储在registryKey:lockKey 这个key下，默认的过期时间是60秒，线程解锁一个已经过期的锁会导致IllegalStateException.这应该被视为一个严重错误，因为它可能会导致被保护的资源被破坏。 然而，锁只在registry中有效，如果使用相同的registryKey,注册在不同的registry中，也会被认为是不同的锁。 注意：这不是为低延迟应用设计的分布式锁实现，这是为在不同JVM之间为资源加锁而设计的实现。

RedisLockRegistry 不仅实现了ExpirableLockRegistry接口，还实现了DisposableBean接口，在spring容器销毁时，可以调用destroy方法来释放资源，具体到这里的资源就是线程池了。

@Override
	public void destroy() {
		if (!this.executorExplicitlySet) {
			((ExecutorService) this.executor).shutdown();
		}
	}

private Executor executor =
			Executors.newCachedThreadPool(new CustomizableThreadFactory("redis-lock-registry-"));
			

该类提供了两个构造方法:

public RedisLockRegistry(RedisConnectionFactory connectionFactory, String registryKey) {
		this(connectionFactory, registryKey, DEFAULT_EXPIRE_AFTER);
	}
public RedisLockRegistry(RedisConnectionFactory connectionFactory, String registryKey, long expireAfter) {
		Assert.notNull(connectionFactory, "'connectionFactory' cannot be null");
		Assert.notNull(registryKey, "'registryKey' cannot be null");
		this.redisTemplate = new StringRedisTemplate(connectionFactory);
		this.obtainLockScript = new DefaultRedisScript<>(OBTAIN_LOCK_SCRIPT, Boolean.class);
		this.registryKey = registryKey;
		this.expireAfter = expireAfter;
	}

构造本类实例，需要提供Redis的连接工厂，registryKey，registryKey的过期时间，过期时间可以使用默认值60秒

private static final long DEFAULT_EXPIRE_AFTER = 60000L;

那么RedisLockRegistry是如何在Redis服务器上设置锁的呢？这就要用到lua脚本的能力了：

private static final String OBTAIN_LOCK_SCRIPT =
			"local lockClientId = redis.call('GET', KEYS[1])\n" +
					"if lockClientId == ARGV[1] then\n" +
					"  redis.call('PEXPIRE', KEYS[1], ARGV[2])\n" +
					"  return true\n" +
					"elseif not lockClientId then\n" +
					"  redis.call('SET', KEYS[1], ARGV[1], 'PX', ARGV[2])\n" +
					"  return true\n" +
					"end\n" +
					"return false";

我们来解读一下这个脚本，首先要调用lua脚本，从Redis中获取KEYS[1] 指定的这个key的值，如果取到的值与ARGV[1]相等，那么就执行PEXPIRE KEYS[1] , ARGV[2];如果没取到值，那么就执行SET KEYS[1] ,ARGV[1],PX ARGV[2]，翻译一下意思就是如果Redis中已经存在了值为ARGV[1]的key,那么就获取到锁，就为该key设置过期时间ARGV[2],如果没有取到值，那么就设置key的值和过期时间。

我们继续看一下如何获取锁:

@Override
	public Lock obtain(Object lockKey) {
		Assert.isInstanceOf(String.class, lockKey);
		String path = (String) lockKey;
		return this.locks.computeIfAbsent(path, RedisLock::new);
	}

obtain 这个方法是在LockRegistry这个接口中定义的方法，locks 是Map<String, RedisLock> locks = new ConcurrentHashMap<>()的实现，ConcurrentHashMap中存放的V是RedisLock的实例，RedisLock是RedisLockRegistry的内部类，实现了Lock接口，内部又持有一个ReentrantLock，也就J.U.C提供的可重入锁的实现，在获取锁时，通过指定一个lockKey，生成指定的RedisLock:

private RedisLock(String path) {
			this.lockKey = constructLockKey(path);
		}

通过源码我们可以发现，最终锁在Redis中的key是registryKey+":"+path这种格式，所以在设置registryKey时，不需要添加后缀的*':'*。

this.locks.computeIfAbsent(path, RedisLock::new)这个方法，如果从locks中无法获取到key为path的值，那么就new 一个RedisLock 并将其放入到locks，如果能取到，那么就返回取到的值，因此，key为path的锁，在当前RedisLockRegistry中只会存在一个，保证了锁的唯一性。

获取到了锁的实例之后，我们还需要尝试申请锁，Lock接口定义了三种申请锁的方法：

• lock()
• tryLock()
• tryLock(long time,TimeUnit unit)

第一种方法会一直尝试获取锁，直到获取到为止，第二种会尝试一次获取锁，获取不到就停止，第三种会在指定时间段内一直尝试，直到获取到锁或时间到期，很显然，我们多数情况下都会使用第三种获取锁的方式。我们看一下RedisLock中重写的第三种方法的实现:

@Override
		public boolean tryLock(long time, TimeUnit unit) throws InterruptedException {
			long now = System.currentTimeMillis();
			if (!this.localLock.tryLock(time, unit)) { ①
				return false;
			}
			try {
				long expire = now + TimeUnit.MILLISECONDS.convert(time, unit);
				boolean acquired;
				while (!(acquired = obtainLock()) && System.currentTimeMillis() < expire) { //NOSONAR ②
					Thread.sleep(100); //NOSONAR ③
				}
				if (!acquired) {
					this.localLock.unlock(); ④
				}
				return acquired;
			}
			catch (Exception e) {
				this.localLock.unlock();
				rethrowAsLockException(e);
			}
			return false;
		}

首先，先RedisLock持有的ReentrantLock尝试获取锁，如果在指定时间范围内无法获取到锁，直接返回false;获取到锁之后，在指定时间范围内，通过obtainLock方法不停尝试在Redis设置锁，每次尝试的间隔是100毫秒，如果在指定时间范围内无法在Redis设置锁，那么需要将内部持有的ReentrantLock释放，如果在Redis成功设置了锁，则返回true，表示获取锁成功。 我们再看一下obtainLock方法的实现：

private boolean obtainLock() {
			Boolean success =
					RedisLockRegistry.this.redisTemplate.execute(RedisLockRegistry.this.obtainLockScript,
							Collections.singletonList(this.lockKey), RedisLockRegistry.this.clientId,
							String.valueOf(RedisLockRegistry.this.expireAfter));

			boolean result = Boolean.TRUE.equals(success);

			if (result) {
				this.lockedAt = System.currentTimeMillis();
			}
			return result;
		}

使用redisTemplate 执行上文提到的OBTAIN_LOCK_SCRIPT这个lua脚本，key的值为RedisLockRegistry.this.clientId，我们可以看到它的实现是UUID.randomUUID().toString();

当前线程使用完资源之后，需要将锁释放，让其他线程有机会获取到锁，我们再看一下释放锁的实现:

@Override
		public void unlock() {
			if (!this.localLock.isHeldByCurrentThread()) { ①
				throw new IllegalStateException("You do not own lock at " + this.lockKey);
			}
			if (this.localLock.getHoldCount() > 1) { ②
				this.localLock.unlock();
				return;
			}
			try {
				if (!isAcquiredInThisProcess()) {
					throw new IllegalStateException("Lock was released in the store due to expiration. " +
							"The integrity of data protected by this lock may have been compromised.");
				}

				if (Thread.currentThread().isInterrupted()) {
					RedisLockRegistry.this.executor.execute(this::removeLockKey);
				}
				else {
					removeLockKey();
				}

				if (LOGGER.isDebugEnabled()) {
					LOGGER.debug("Released lock; " + this);
				}
			}
			catch (Exception e) {
				ReflectionUtils.rethrowRuntimeException(e);
			}
			finally {
				this.localLock.unlock();
			}
		}

首先要判断当前要释放锁的线程是否为锁的加锁线程，释放锁与加锁线程必须为同一线程才能释放锁，第二步要判断当前线程持有几个锁（Queries the number of holds on this lock by the current thread.），因为ReentrantLock是可重入锁，如果持有个数大于1，说明当前锁被持有了多次，不能将Redis中的锁释放，只能是将RedisLock中持有的ReentrantLock的持有数减1，如果持有数为1，那么就需要释放Redis锁：

private void removeLockKey() {
			if (RedisLockRegistry.this.unlinkAvailable) {
				try {
					RedisLockRegistry.this.redisTemplate.unlink(this.lockKey);
				}
				catch (Exception ex) {
					RedisLockRegistry.this.unlinkAvailable = false;
					if (LOGGER.isDebugEnabled()) {
						LOGGER.debug("The UNLINK command has failed (not supported on the Redis server?); " +
								"falling back to the regular DELETE command", ex);
					}
					else {
						LOGGER.warn("The UNLINK command has failed (not supported on the Redis server?); " +
								"falling back to the regular DELETE command: " + ex.getMessage());
					}
					RedisLockRegistry.this.redisTemplate.delete(this.lockKey);
				}
			}
			else {
				RedisLockRegistry.this.redisTemplate.delete(this.lockKey);
			}
		}

分析完上锁的流程，释放锁的流程就很好理解了。

文章开头提到了分布式锁的特性中还包括响应中断，大家应该都有了解，线程是否能响应中断，基本需要看造化，我个人不建议在程序中使用中断，程序执行到哪一步，谁能说的清呢，所以这里就不展开说了。

Redis的Java客户端，除了上文提到过的Lettuce和Jedis,还有一个Redisson，这个客户端也非常不错，后面有时间我们再看看Redisson的分布式锁是如何实现的。

另外，为了提高Redis的高可用性，一般都会使用主从模式部署，那么如果master宕机了，Redis中的数据需要异步同步到从节点，从节点需要一边同步数据，一边为客户端提供服务，那么这时候如果锁数据还未同步完成，新的客户端申请锁，就会导致两个客户端同时持有了相同的锁，所以，这种分布式锁的实现方式，不能100%的实现排他性，那要怎么设计分布式锁呢？

其实Redis作者已经给出了答案，在Redis官网上，作者介绍了Redlock，但是Redlock的实现比较复杂，之后我们再详细介绍。