基於 Redis 實現分布式應用限流[轉]

限流的目的是通過對並發訪問/請求進行限速或者一個時間窗口內的的請求進行限速來保護系統，一旦達到限制速率則可以拒絕服務。

前幾天在DD的公眾號，看了一篇關於使用 瓜娃 實現單應用限流的方案，參考《redis in action》 實現了一個jedis版本的，都屬於業務層次限制。 實際場景中常用的限流策略：

• Nginx接入層限流
  按照一定的規則如帳號、IP、系統調用邏輯等在Nginx層面做限流
• 業務應用系統限流
  通過業務代碼控制流量這個流量可以被稱為信號量，可以理解成是一種鎖，它可以限制一項資源最多能同時被多少進程訪問。

  代碼實現

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

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

  /**
  * email [e-mail]
  * @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();

  //刪除信號量
  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);

  //計數器自增
  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();
  //獲取排名，判斷請求是否取得了信號量
  long rank = (Long) results.get(results.size() - 1);
  if (rank < limit) {
  return identifier;
  } else {//沒有獲取到信號量，清理之前放入redis 中垃圾數據
  transaction = jedis.multi();
  transaction.zrem(BUCKET_MONITOR, identifier);
  transaction.zrem(BUCKET, identifier);
  transaction.exec();
  }
  return null;
  }
  }

• 調用

   

   

  測試接口調用 @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 你的業務邏輯 } jedisPool.returnResource(jedis); }

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

  測試接口調用 @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 你的業務邏輯     }     jedisPool.returnResource(jedis); }

   

  優化

  使用攔截器 + 注解優化代碼

  攔截器

   

   

  @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("/*");     } }

   

  定義注解

   

   

  /** * email [e-mail] * @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 [e-mail] * @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() { }

   

  並發測試

  工具：apache-jmeter-3.2
  說明： 沒有獲取到信號量的接口返回500，status是紅色，獲取到信號量的接口返回200，status是綠色。
  當限制請求信號量為2，並發5個線程：
  當限制請求信號量為5，並發10個線程：
  

  資料

  基於reids + lua的實現

  總結

  1. 對於信號量的操作，使用事務操作。
  2. 不要使用時間戳作為信號量的排序分數，因為在分布式環境中，各個節點的時間差的原因，會出現不公平信號量的現象。
  3. 可以使用把這塊代碼抽成@rateLimiter注解，然后再方法上使用就會很方便啦
  4. 不同接口的流控，可以參考源碼的里面RedisRateLimiterPlus,無非是每個接口生成一個監控參數
  5. 源碼http://git.oschina.net/boding1/pig-cloud