分布式锁大有用途,比如用在减库存操作、流水号生成,分布式计数器等。分布式锁服务在大家的项目中或许用的不多,因为大家都把排他放在数据库那一层来挡。当大量的行锁、表锁、事务充斥着数据库的时候。一般web应用很多的瓶颈都在数据库上,这里给大家介绍的是减轻数据库锁负担的方案--分布式锁服务。本文介绍分布式锁常用的三种实现方式。
一、zookeeper
1、实现原理:
基于zookeeper瞬时有序节点实现的分布式锁,其主要逻辑如下(该图来自于IBM网站)。大致思想即为:每个客户端对某个功能加锁时,在zookeeper上的与该功能对应的指定节点的目录下,生成一个唯一的瞬时有序节点。判断是否获取锁的方式很简单,只需要判断有序节点中序号最小的一个。当释放锁的时候,只需将这个瞬时节点删除即可。同时,其可以避免服务宕机导致的锁无法释放,而产生的死锁问题。
2、优点
锁安全性高,zk可持久化
3、缺点
性能开销比较高。因为其需要动态产生、销毁瞬时节点来实现锁功能。
4、实现
可以直接采用zookeeper第三方库curator即可方便地实现分布式锁。以下为基于curator实现的zk分布式锁核心代码:
- @Override
- public boolean tryLock(LockInfo info) {
- InterProcessMutex mutex = getMutex(info);
- int tryTimes = info.getTryTimes();
- long tryInterval = info.getTryInterval();
- boolean flag = true;// 代表是否需要重试
- while (flag && --tryTimes >= 0) {
- try {
- if (mutex.acquire(info.getWaitLockTime(), TimeUnit.MILLISECONDS)) {
- LOGGER.info(LogConstant.DST_LOCK + "acquire lock successfully!");
- flag = false;
- break;
- }
- } catch (Exception e) {
- LOGGER.error(LogConstant.DST_LOCK + "acquire lock error!", e);
- } finally {
- checkAndRetry(flag, tryInterval, tryTimes);
- }
- }
- return !flag;// 最后还需要重试,说明没拿到锁
- }
- @Override
- public boolean releaseLock(LockInfo info) {
- InterProcessMutex mutex = getMutex(info);
- int tryTimes = info.getTryTimes();
- long tryInterval = info.getTryInterval();
- boolean flag = true;// 代表是否需要重试
- while (flag && --tryTimes >= 0) {
- try {
- mutex.release();
- LOGGER.info(LogConstant.DST_LOCK + "release lock successfully!");
- flag = false;
- break;
- } catch (Exception e) {
- LOGGER.error(LogConstant.DST_LOCK + "release lock error!", e);
- } finally {
- checkAndRetry(flag, tryInterval, tryTimes);
- }
- }
- return !flag;// 最后还需要重试,说明没拿到锁
- }
- /**
- * 获取锁。此处需要加同步,concurrentHashmap无法避免此处的同步问题
- * @param info 锁信息
- * @return 锁实例
- */
- private synchronized InterProcessMutex getMutex(LockInfo info) {
- InterProcessReadWriteLock lock = null;
- if (locksCache.get(info.getLock()) != null) {
- lock = locksCache.get(info.getLock());
- } else {
- lock = new InterProcessReadWriteLock(client, BASE_DIR + info.getLock());
- locksCache.put(info.getLock(), lock);
- }
- InterProcessMutex mutex = null;
- switch (info.getIsolate()) {
- case READ:
- mutex = lock.readLock();
- break;
- case WRITE:
- mutex = lock.writeLock();
- break;
- default:
- throw new IllegalArgumentException();
- }
- return mutex;
- }
- /**
- * 判断是否需要重试
- * @param flag 是否需要重试标志
- * @param tryInterval 重试间隔
- * @param tryTimes 重试次数
- */
- private void checkAndRetry(boolean flag, long tryInterval, int tryTimes) {
- try {
- if (flag) {
- Thread.sleep(tryInterval);
- LOGGER.info(LogConstant.DST_LOCK + "retry getting lock! now retry time left: " + tryTimes);
- }
- } catch (InterruptedException e) {
- LOGGER.error(LogConstant.DST_LOCK + "retry interval thread interruptted!", e);
- }
- }
二、memcached分布式锁
1、实现原理:
memcached带有add函数,利用add函数的特性即可实现分布式锁。add和set的区别在于:如果多线程并发set,则每个set都会成功,但最后存储的值以最后的set的线程为准。而add的话则相反,add会添加第一个到达的值,并返回true,后续的添加则都会返回false。利用该点即可很轻松地实现分布式锁。
2、优点
并发高效。
3、缺点
(1)memcached采用列入LRU置换策略,所以如果内存不够,可能导致缓存中的锁信息丢失。
(2)memcached无法持久化,一旦重启,将导致信息丢失。
三、redis分布式锁
redis分布式锁即可以结合zk分布式锁锁高度安全和memcached并发场景下效率很好的优点,可以利用jedis客户端实现。
/** * @author http://blog.csdn.net/java2000_wl * @version 1.0.0 */public class RedisBillLockHandler implements IBatchBillLockHandler { private static final Logger LOGGER = LoggerFactory.getLogger(RedisBillLockHandler.class); private static final int DEFAULT_SINGLE_EXPIRE_TIME = 3; private static final int DEFAULT_BATCH_EXPIRE_TIME = 6; private final JedisPool jedisPool; /** * 构造 * @author http://blog.csdn.net/java2000_wl */ public RedisBillLockHandler(JedisPool jedisPool) { this.jedisPool = jedisPool; } /** * 获取锁 如果锁可用 立即返回true, 否则返回false * @author http://blog.csdn.net/java2000_wl * @param billIdentify * @return */ public boolean tryLock(IBillIdentify billIdentify) { return tryLock(billIdentify, 0L, null); } /** * 锁在给定的等待时间内空闲,则获取锁成功 返回true, 否则返回false * @author http://blog.csdn.net/java2000_wl * @param billIdentify * @param timeout * @param unit * @return */ public boolean tryLock(IBillIdentify billIdentify, long timeout, TimeUnit unit) { String key = (String) billIdentify.uniqueIdentify(); Jedis jedis = null; try { jedis = getResource(); long nano = System.nanoTime(); do { LOGGER.debug("try lock key: " + key); Long i = jedis.setnx(key, key); if (i == 1) { jedis.expire(key, DEFAULT_SINGLE_EXPIRE_TIME); LOGGER.debug("get lock, key: " + key + " , expire in " + DEFAULT_SINGLE_EXPIRE_TIME + " seconds."); return Boolean.TRUE; } else { // 存在锁 if (LOGGER.isDebugEnabled()) { String desc = jedis.get(key); LOGGER.debug("key: " + key + " locked by another business:" + desc); } } if (timeout == 0) { break; } Thread.sleep(300); } while ((System.nanoTime() - nano) < unit.toNanos(timeout)); return Boolean.FALSE; } catch (JedisConnectionException je) { LOGGER.error(je.getMessage(), je); returnBrokenResource(jedis); } catch (Exception e) { LOGGER.error(e.getMessage(), e); } finally { returnResource(jedis); } return Boolean.FALSE; } /** * 如果锁空闲立即返回 获取失败 一直等待 * @author http://blog.csdn.net/java2000_wl * @param billIdentify */ public void lock(IBillIdentify billIdentify) { String key = (String) billIdentify.uniqueIdentify(); Jedis jedis = null; try { jedis = getResource(); do { LOGGER.debug("lock key: " + key); Long i = jedis.setnx(key, key); if (i == 1) { jedis.expire(key, DEFAULT_SINGLE_EXPIRE_TIME); LOGGER.debug("get lock, key: " + key + " , expire in " + DEFAULT_SINGLE_EXPIRE_TIME + " seconds."); return; } else { if (LOGGER.isDebugEnabled()) { String desc = jedis.get(key); LOGGER.debug("key: " + key + " locked by another business:" + desc); } } Thread.sleep(300); } while (true); } catch (JedisConnectionException je) { LOGGER.error(je.getMessage(), je); returnBrokenResource(jedis); } catch (Exception e) { LOGGER.error(e.getMessage(), e); } finally { returnResource(jedis); } } /** * 释放锁 * @author http://blog.csdn.net/java2000_wl * @param billIdentify */ public void unLock(IBillIdentify billIdentify) { Listlist = new ArrayList (); list.add(billIdentify); unLock(list); } /** * 批量获取锁 如果全部获取 立即返回true, 部分获取失败 返回false * @author http://blog.csdn.net/java2000_wl * @date 2013-7-22 下午10:27:44 * @param billIdentifyList * @return */ public boolean tryLock(List billIdentifyList) { return tryLock(billIdentifyList, 0L, null); } /** * 锁在给定的等待时间内空闲,则获取锁成功 返回true, 否则返回false * @author http://blog.csdn.net/java2000_wl * @param billIdentifyList * @param timeout * @param unit * @return */ public boolean tryLock(List billIdentifyList, long timeout, TimeUnit unit) { Jedis jedis = null; try { List needLocking = new CopyOnWriteArrayList (); List locked = new CopyOnWriteArrayList (); jedis = getResource(); long nano = System.nanoTime(); do { // 构建pipeline,批量提交 Pipeline pipeline = jedis.pipelined(); for (IBillIdentify identify : billIdentifyList) { String key = (String) identify.uniqueIdentify(); needLocking.add(key); pipeline.setnx(key, key); } LOGGER.debug("try lock keys: " + needLocking); // 提交redis执行计数 List