基于 Redis 实现分布式应用限流[转]

限流的目的是通过对并发访问/请求进行限速或者一个时间窗口内的的请求进行限速来保护系统，一旦达到限制速率则可以拒绝服务。

前几天在DD的公众号，看了一篇关于使用 瓜娃 实现单应用限流的方案，参考《redis in action》 实现了一个jedis版本的，都属于业务层次限制。 实际场景中常用的限流策略：

• Nginx接入层限流
  按照一定的规则如帐号、IP、系统调用逻辑等在Nginx层面做限流
• 业务应用系统限流
  通过业务代码控制流量这个流量可以被称为信号量，可以理解成是一种锁，它可以限制一项资源最多能同时被多少进程访问。

  代码实现

  import redis.clients.jedis.Jedis;
  import redis.clients.jedis.Transaction;
  import redis.clients.jedis.ZParams;

  import java.util.List;
  import java.util.UUID;

  /**
  * email wangiegie@gmail.com
  * @data 2017-08
  */
  public class RedisRateLimiter {
  private static final String BUCKET = "BUCKET";
  private static final String BUCKET_COUNT = "BUCKET_COUNT";
  private static final String BUCKET_MONITOR = "BUCKET_MONITOR";

  static String acquireTokenFromBucket(
  Jedis jedis, int limit, long timeout) {
  String identifier = UUID.randomUUID().toString();
  long now = System.currentTimeMillis();
  Transaction transaction = jedis.multi();

  //删除信号量
  transaction.zremrangeByScore(BUCKET_MONITOR.getBytes(), "-inf".getBytes(), String.valueOf(now - timeout).getBytes());
  ZParams params = new ZParams();
  params.weightsByDouble(1.0,0.0);
  transaction.zinterstore(BUCKET, params, BUCKET, BUCKET_MONITOR);

  //计数器自增
  transaction.incr(BUCKET_COUNT);
  List<Object> results = transaction.exec();
  long counter = (Long) results.get(results.size() - 1);

  transaction = jedis.multi();
  transaction.zadd(BUCKET_MONITOR, now, identifier);
  transaction.zadd(BUCKET, counter, identifier);
  transaction.zrank(BUCKET, identifier);
  results = transaction.exec();
  //获取排名，判断请求是否取得了信号量
  long rank = (Long) results.get(results.size() - 1);
  if (rank < limit) {
  return identifier;
  } else {//没有获取到信号量，清理之前放入redis 中垃圾数据
  transaction = jedis.multi();
  transaction.zrem(BUCKET_MONITOR, identifier);
  transaction.zrem(BUCKET, identifier);
  transaction.exec();
  }
  return null;
  }
  }

• 调用

   

   

  测试接口调用 @GetMapping("/") public void index(HttpServletResponse response) throws IOException { Jedis jedis = jedisPool.getResource(); String token = RedisRateLimiter.acquireTokenFromBucket(jedis, LIMIT, TIMEOUT); if (token == null) { response.sendError(500); }else{ //TODO 你的业务逻辑 } jedisPool.returnResource(jedis); }

  1 2 3 4 5 6 7 8 9 10 11 12 13

  测试接口调用 @GetMapping("/") public void index(HttpServletResponse response) throws IOException {     Jedis jedis = jedisPool.getResource();     String token = RedisRateLimiter.acquireTokenFromBucket(jedis, LIMIT, TIMEOUT);     if (token == null) {         response.sendError(500);     }else{         //TODO 你的业务逻辑     }     jedisPool.returnResource(jedis); }

   

  优化

  使用拦截器 + 注解优化代码

  拦截器

   

   

  @Configuration static class WebMvcConfigurer extends WebMvcConfigurerAdapter { private Logger logger = LoggerFactory.getLogger(WebMvcConfigurer.class); @Autowired private JedisPool jedisPool; public void addInterceptors(InterceptorRegistry registry) { registry.addInterceptor(new HandlerInterceptorAdapter() { public boolean preHandle(HttpServletRequest request, HttpServletResponse response, Object handler) throws Exception { HandlerMethod handlerMethod = (HandlerMethod) handler; Method method = handlerMethod.getMethod(); RateLimiter rateLimiter = method.getAnnotation(RateLimiter.class); if (rateLimiter != null){ int limit = rateLimiter.limit(); int timeout = rateLimiter.timeout(); Jedis jedis = jedisPool.getResource(); String token = RedisRateLimiter.acquireTokenFromBucket(jedis, limit, timeout); if (token == null) { response.sendError(500); return false; } logger.debug("token -> {}",token); jedis.close(); } return true; } }).addPathPatterns("/*"); } }

  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32

  @Configuration static class WebMvcConfigurer extends WebMvcConfigurerAdapter {     private Logger logger = LoggerFactory.getLogger(WebMvcConfigurer.class);     @Autowired     private JedisPool jedisPool;       public void addInterceptors(InterceptorRegistry registry) {         registry.addInterceptor(new HandlerInterceptorAdapter() {             public boolean preHandle(HttpServletRequest request, HttpServletResponse response,                                      Object handler) throws Exception {                 HandlerMethod handlerMethod = (HandlerMethod) handler;                 Method method = handlerMethod.getMethod();                 RateLimiter rateLimiter = method.getAnnotation(RateLimiter.class);                   if (rateLimiter != null){                     int limit = rateLimiter.limit();                     int timeout = rateLimiter.timeout();                     Jedis jedis = jedisPool.getResource();                     String token = RedisRateLimiter.acquireTokenFromBucket(jedis, limit, timeout);                     if (token == null) {                         response.sendError(500);                         return false;                     }                     logger.debug("token -> {}",token);                     jedis.close();                 }                 return true;             }         }).addPathPatterns("/*");     } }

   

  定义注解

   

   

  /** * email wangiegie@gmail.com * @data 2017-08 * 限流注解 */ @Target(ElementType.METHOD) @Retention(RetentionPolicy.RUNTIME) @Documented public @interface RateLimiter { int limit() default 5; int timeout() default 1000; }

  1 2 3 4 5 6 7 8 9 10 11 12 13 14

  /** * email wangiegie@gmail.com * @data 2017-08 * 限流注解 */   @Target(ElementType.METHOD) @Retention(RetentionPolicy.RUNTIME) @Documented public @interface RateLimiter {     int limit() default 5;     int timeout() default 1000; }

   

  使用

   

   

  @RateLimiter(limit = 2, timeout = 5000) @GetMapping("/test") public void test() { }

  1 2 3 4 5

  @RateLimiter(limit = 2, timeout = 5000) @GetMapping("/test") public void test() { }

   

  并发测试

  工具：apache-jmeter-3.2
  说明： 没有获取到信号量的接口返回500，status是红色，获取到信号量的接口返回200，status是绿色。
  当限制请求信号量为2，并发5个线程：
  当限制请求信号量为5，并发10个线程：
  

  资料

  基于reids + lua的实现

  总结

  1. 对于信号量的操作，使用事务操作。
  2. 不要使用时间戳作为信号量的排序分数，因为在分布式环境中，各个节点的时间差的原因，会出现不公平信号量的现象。
  3. 可以使用把这块代码抽成@rateLimiter注解，然后再方法上使用就会很方便啦
  4. 不同接口的流控，可以参考源码的里面RedisRateLimiterPlus,无非是每个接口生成一个监控参数
  5. 源码http://git.oschina.net/boding1/pig-cloud