Spring Cloud 使用 Resilience4j 实现服务熔断

CircuitBreaker 断路器

服务熔断是为了保护我们的服务，比如当某个服务出现问题的时候，控制打向它的流量，让它有时间去恢复，或者限制一段时间只能有固定数量的请求打向这个服务。这些都是保护措施。我在实际工作中也确实遇到过，数据库出现问题了，进而导致Web服务出现问题了，导致不依赖数据库的服务也出现问题了，出现一连串问题。 这次学习《玩转 Spring 全家桶》，丁雪丰老师给了使用resilience4j的例子。 丁老师的例子是2019年的，这个框架已经修改了些方法，所以我自己也花了些时间来理解了它的用法。现将过程记录下来。
首先POM文件引入

1. <dependency>
2.     <groupId>io.github.resilience4j</groupId>
3.     <artifactId>resilience4j-spring-boot2</artifactId>
4.     <version>2.0.2</version>
5. </dependency>

复制代码

接着改造之前的Controller方法

1. @RestController
2. @RequestMapping("/customer")
3. @Slf4j
4. public class BookController {
6.     @Autowired
7.     private BookService bookService;
9.     private CircuitBreaker circuitBreaker;
11.     public BookController(CircuitBreakerRegistry registry) {
12.         circuitBreaker = registry.circuitBreaker("menu");
13.     }
15.     @GetMapping("/menu")
16.     public List<Book> readMenu() {
17.         Supplier<List<Book>> supplier = () -> bookService.getAll();
18.         circuitBreaker.getEventPublisher()
19.                 .onEvent(event -> log.info(event.toString()));
20.         try{
21.             return circuitBreaker.executeSupplier(supplier);
22.         }
23.         catch (Exception ex)
24.         {
25.             log.error(ex.getMessage());
26.             return Collections.emptyList();
27.         }
28.     }
29. }

复制代码

不同的地方就是引入了CircuitBreaker， 然后使用它将我们的方法“bookService.getAll()”包起来了。
然后在配置文件中添加如下的配置

1. resilience4j.circuitbreaker.backends.menu.failure-rate-threshold=50
2. resilience4j.circuitbreaker.backends.menu.wait-duration-in-open-state=60000
3. resilience4j.circuitbreaker.backends.menu.sliding-window-size=5
4. resilience4j.circuitbreaker.backends.menu.permitted-number-of-calls-in-half-open-state=2
5. resilience4j.circuitbreaker.backends.menu.minimum-number-of-calls=2

复制代码

稍微解释一下这里的配置
failure-rate-threshold=50是说失败率超过50%就熔断，
wait-duration-in-open-state= 60000，是说熔断后等待60S才允许再次调用。
sliding-window-size =5 可以理解为5个请求统计一次，
permitted-number-of-calls-in-half-open-state = 2是说进入半开的状态的时候，还允许请求多少个。
minimum-number-of-calls=2是说最少有多少个请求才开始统计。 这里的参数都是我为了实验设置的，实际情况根据需要进行调整。参数比较多，具体可以参加官方文档
https://resilience4j.readme.io/docs/circuitbreaker
我们来看下实际的效果通过浏览器访问，
首先我们现打开BookService，让它有一次成功的请求，日志会输出
CircuitBreaker 'menu' recorded a successful call.
然后我们将BookService关闭，让它请求失败，日志会输出如下
CircuitBreaker 'menu' recorded an error: 'feign.RetryableException: Connection refused: no further information executing GET http://bookshop-service/book/getAll'. Elapsed time: 2050 ms
CircuitBreaker 'menu' exceeded failure rate threshold. Current failure rate: 50.0
CircuitBreaker 'menu' changed state from CLOSED to OPEN
可以看到断路器已经打开了，
接着我们继续访问会出现，
CircuitBreaker 'menu' recorded a call which was not permitted.
这个时候请求不会打到BookService上面了。就算这个时候我们的BookService恢复正常。
等待60s后进入半Open的状态
CircuitBreaker 'menu' changed state from OPEN to HALF_OPEN
这个时候恢复BookService正常，我们请求也会正常响应了
CircuitBreaker 'menu' recorded a successful call
多请求几次，断路器就从HALF_OPEN变成了CLOSED
CircuitBreaker 'menu' changed state from HALF_OPEN to CLOSED
这里给一个官方的状态图来说明

断路器有三个状态： CLOSED，  OPEN， HALF_OPEN。

• CLOSED是最开始的状态，也就是关闭状态，流量可以正常通过，
  
• 当失败比率超过threshold后，断路器打开， 变成OPEN 打开后流量不可以通过；
  
• 等待一定的时间后，断路器进入半开状态 HALF_OPEN, 这个时候如果失败率低于阈值，断路器进入CLOSED状态，如果超过阈值，断路器继续保证OPEN，再等待，如此往复。
  

断路器现在还支持设置慢请求，使用起来还是比较方便。对于参数的设置如果不是很理解，可以通过单元测试的方法来加深对它的理解。这里参考https://github.com/eugenp/tutorials/blob/master/libraries-6/src/test/java/com/baeldung/resilence4j/Resilience4jUnitTest.java 上面的例子，给出来个单元测试

1.     interface RemoteService {
3.         int process(int i);
4.     }
6.     private RemoteService service;
9.     @Test
10.     public void whenCircuitBreakerIsUsed_thenItWorksAsExpected() {
11.         service = mock(RemoteService.class);
12.         CircuitBreakerConfig config = CircuitBreakerConfig.custom()
13.                 // Percentage of failures to start short-circuit
14.                 .failureRateThreshold(20)
15.                 .minimumNumberOfCalls(5)
16.                 .build();
17.         CircuitBreakerRegistry registry = CircuitBreakerRegistry.of(config);
18.         CircuitBreaker circuitBreaker = registry.circuitBreaker("my");
19.         Function<Integer, Integer> decorated = CircuitBreaker.decorateFunction(circuitBreaker, service::process);
21.         when(service.process(anyInt())).thenThrow(new RuntimeException());
22.         circuitBreaker.getEventPublisher()
23.                 .onEvent(event ->
24.                 {
25.                     log.info(event.toString());
26.                 });
28.         for (int i = 0; i < 10; i++) {
29.             try {
30.                 decorated.apply(i);
31.             } catch (Exception ignore) {
32.             }
33.         }
35.         verify(service, times(5)).process(any(Integer.class));
36.     }

复制代码

这里设置最少请求5次，失败率超过20%就熔断，然后我们请求了10次，实际上只调用了Service5次。
对于其它参数，你可以调整后，根据需要来验证是否符合预期。它的日志输出如下

1. CircuitBreaker 'my' recorded an error: 'java.lang.RuntimeException'. Elapsed time: 2 ms
2. CircuitBreaker 'my' recorded an error: 'java.lang.RuntimeException'. Elapsed time: 0 ms
3. CircuitBreaker 'my' recorded an error: 'java.lang.RuntimeException'. Elapsed time: 0 ms
4. CircuitBreaker 'my' recorded an error: 'java.lang.RuntimeException'. Elapsed time: 0 ms
5. CircuitBreaker 'my' recorded an error: 'java.lang.RuntimeException'. Elapsed time: 0 ms
6. CircuitBreaker 'my' exceeded failure rate threshold. Current failure rate: 100.0
7. CircuitBreaker 'my' changed state from CLOSED to OPEN
8. CircuitBreaker 'my' recorded a call which was not permitted.
9. CircuitBreaker 'my' recorded a call which was not permitted.
10. CircuitBreaker 'my' recorded a call which was not permitted.
11. CircuitBreaker 'my' recorded a call which was not permitted.
12. CircuitBreaker 'my' recorded a call which was not permitted.  

复制代码

可以看到5次过后，就开始打开断路器，后面的call就不被允许了。
隔舱Bulkhead

Resilience4j 里面的Bulkhead可以简单的理解为允许多少个并发访问。我们这里还是通过单元测试的方法来演示它的功能

1.     @Test
2.     public void whenBulkheadIsUsed_thenItWorksAsExpected() throws InterruptedException {
3.         service = mock(RemoteService.class);
4.         BulkheadConfig config = BulkheadConfig.custom().maxConcurrentCalls(2).build();
5.         BulkheadRegistry registry = BulkheadRegistry.of(config);
6.         Bulkhead bulkhead = registry.bulkhead("my");
7.         Function<Integer, Integer> decorated = Bulkhead.decorateFunction(bulkhead, service::process);
9.        try {
10.             callAndBlock(decorated);
11.         }
12.        catch(BulkheadFullException ex)
13.         {
14.             log.error("isfull");
15.         }
16.         finally
17.         {
18.            verify(service, times(2)).process(any(Integer.class));
19.         }
21.     }
23.     private void callAndBlock(Function<Integer, Integer> decoratedService) throws InterruptedException {
24.         when(service.process(anyInt())).thenAnswer(invocation -> {
25.             log.info("service called");
26.             return null;
27.         });
29.         ArrayList<Integer> numberList = new ArrayList<Integer>();
30.         for(int i = 0;i<10;i++)
31.         {
32.             numberList.add(i);
33.         }
35.         numberList.parallelStream().forEach((i)->{
36.             try {
37.                 decoratedService.apply(i);
38.             }
39.             catch (Exception ex)
40.             {
41.                 log.error("meet error " + ex.getMessage());
42.             } catch (Throwable e) {
43.                 throw new RuntimeException(e);
44.             }
45.         });
46.     }

复制代码

首先我们解读一下callAndBlock， 它会并发的去执行一个function. 如果我们不用隔舱，它的输出会是这样。

1. 2022-12-28T15:22:52.010+08:00  INFO 37276 --- [onPool-worker-4] c.k.r.bookcustomer.Resilience4jUnitTest  : service called
2. 2022-12-28T15:22:52.010+08:00  INFO 37276 --- [onPool-worker-9] c.k.r.bookcustomer.Resilience4jUnitTest  : service called
3. 2022-12-28T15:22:52.010+08:00  INFO 37276 --- [onPool-worker-5] c.k.r.bookcustomer.Resilience4jUnitTest  : service called
4. 2022-12-28T15:22:52.010+08:00  INFO 37276 --- [onPool-worker-3] c.k.r.bookcustomer.Resilience4jUnitTest  : service called
5. 2022-12-28T15:22:52.010+08:00  INFO 37276 --- [onPool-worker-1] c.k.r.bookcustomer.Resilience4jUnitTest  : service called
6. 2022-12-28T15:22:52.010+08:00  INFO 37276 --- [onPool-worker-6] c.k.r.bookcustomer.Resilience4jUnitTest  : service called
7. 2022-12-28T15:22:52.010+08:00  INFO 37276 --- [onPool-worker-7] c.k.r.bookcustomer.Resilience4jUnitTest  : service called
8. 2022-12-28T15:22:52.010+08:00  INFO 37276 --- [onPool-worker-8] c.k.r.bookcustomer.Resilience4jUnitTest  : service called
9. 2022-12-28T15:22:52.010+08:00  INFO 37276 --- [onPool-worker-2] c.k.r.bookcustomer.Resilience4jUnitTest  : service called
10. 2022-12-28T15:22:52.011+08:00  INFO 37276 --- [           main] c.k.r.bookcustomer.Resilience4jUnitTest  : service called

复制代码

可以看到启动了10个线程去访问方法。加了隔舱后，隔舱限定了一次只能两个，输出如下

1. 2022-12-28T15:33:48.648+08:00 ERROR 32256 --- [onPool-worker-4] c.k.r.bookcustomer.Resilience4jUnitTest  : meet error Bulkhead 'my' is full and does not permit further calls
2. 2022-12-28T15:33:48.648+08:00 ERROR 32256 --- [onPool-worker-6] c.k.r.bookcustomer.Resilience4jUnitTest  : meet error Bulkhead 'my' is full and does not permit further calls
3. 2022-12-28T15:33:48.648+08:00 ERROR 32256 --- [onPool-worker-7] c.k.r.bookcustomer.Resilience4jUnitTest  : meet error Bulkhead 'my' is full and does not permit further calls
4. 2022-12-28T15:33:48.648+08:00 ERROR 32256 --- [onPool-worker-4] c.k.r.bookcustomer.Resilience4jUnitTest  : meet error Bulkhead 'my' is full and does not permit further calls
5. 2022-12-28T15:33:48.648+08:00 ERROR 32256 --- [onPool-worker-8] c.k.r.bookcustomer.Resilience4jUnitTest  : meet error Bulkhead 'my' is full and does not permit further calls
6. 2022-12-28T15:33:48.648+08:00 ERROR 32256 --- [onPool-worker-5] c.k.r.bookcustomer.Resilience4jUnitTest  : meet error Bulkhead 'my' is full and does not permit further calls
7. 2022-12-28T15:33:48.648+08:00 ERROR 32256 --- [onPool-worker-2] c.k.r.bookcustomer.Resilience4jUnitTest  : meet error Bulkhead 'my' is full and does not permit further calls
8. 2022-12-28T15:33:48.648+08:00 ERROR 32256 --- [onPool-worker-3] c.k.r.bookcustomer.Resilience4jUnitTest  : meet error Bulkhead 'my' is full and does not permit further calls
9. 2022-12-28T15:33:48.650+08:00  INFO 32256 --- [onPool-worker-1] c.k.r.bookcustomer.Resilience4jUnitTest  : service called
10. 2022-12-28T15:33:48.650+08:00  INFO 32256 --- [           main] c.k.r.bookcustomer.Resilience4jUnitTest  : service called

复制代码

可以看到只有两次成功的访问，其它的访问都被block了。
限速器RateLimiter

RateLimiter的功能是限定一段时间内允许多少次访问，还是使用和Bulkhead一样的例子一样

1. @Test
2.     public void whenRateLimiterInUse_thenItWorksAsExpected() throws InterruptedException {
3.         service = mock(RemoteService.class);
5.         RateLimiterConfig config = RateLimiterConfig.custom()
6.                 .limitRefreshPeriod(Duration.ofMillis(1000))
7.                 .limitForPeriod(4)
8.                 .timeoutDuration(Duration.ofMillis(25))
9.                 .build();
11.         RateLimiterRegistry rateLimiterRegistry = RateLimiterRegistry.of(config);
13.         RateLimiter rateLimiter = rateLimiterRegistry
14.                 .rateLimiter("name1");
16.         CheckedFunction<Integer, Integer> decorated = RateLimiter
17.                 .decorateCheckedFunction(rateLimiter, service::process);
19.         try {
20.             callAndBlock(decorated);
21.         }
22.         catch(Exception ex)
23.         {
24.             log.error("isfull");
25.         }
26.         finally
27.         {
28.             verify(service, times(4)).process(any(Integer.class));
29.         }
31.     }
33.     private void callAndBlock(CheckedFunction<Integer, Integer> decoratedService) throws InterruptedException {
34.         when(service.process(anyInt())).thenAnswer(invocation -> {
35.             log.info("service called");
36.             return null;
37.         });
39.         ArrayList<Integer> numberList = new ArrayList<Integer>();
40.         for(int i = 0;i<10;i++)
41.         {
42.             numberList.add(i);
43.         }
45.         numberList.parallelStream().forEach((i)->{
46.             try {
47.                 decoratedService.apply(i);
48.             }
49.             catch (Exception ex)
50.             {
51.                 log.error("meet error " + ex.getMessage());
52.             } catch (Throwable e) {
53.                 throw new RuntimeException(e);
54.             }
55.         });
56.     }

复制代码

我们这里故意设置1S中允许访问4次，实际的运行情况也是只允许了4次。日志输出如下

1. 2022-12-28T15:39:52.027+08:00  INFO 35236 --- [onPool-worker-2] c.k.r.bookcustomer.Resilience4jUnitTest  : service called
2. 2022-12-28T15:39:52.027+08:00  INFO 35236 --- [onPool-worker-5] c.k.r.bookcustomer.Resilience4jUnitTest  : service called
3. 2022-12-28T15:39:52.027+08:00  INFO 35236 --- [           main] c.k.r.bookcustomer.Resilience4jUnitTest  : service called
4. 2022-12-28T15:39:52.027+08:00  INFO 35236 --- [onPool-worker-7] c.k.r.bookcustomer.Resilience4jUnitTest  : service called
5. 2022-12-28T15:39:52.053+08:00 ERROR 35236 --- [onPool-worker-6] c.k.r.bookcustomer.Resilience4jUnitTest  : meet error RateLimiter 'name1' does not permit further calls
6. 2022-12-28T15:39:52.060+08:00 ERROR 35236 --- [onPool-worker-3] c.k.r.bookcustomer.Resilience4jUnitTest  : meet error RateLimiter 'name1' does not permit further calls
7. 2022-12-28T15:39:52.060+08:00 ERROR 35236 --- [onPool-worker-9] c.k.r.bookcustomer.Resilience4jUnitTest  : meet error RateLimiter 'name1' does not permit further calls
8. 2022-12-28T15:39:52.060+08:00 ERROR 35236 --- [onPool-worker-1] c.k.r.bookcustomer.Resilience4jUnitTest  : meet error RateLimiter 'name1' does not permit further calls
9. 2022-12-28T15:39:52.060+08:00 ERROR 35236 --- [onPool-worker-8] c.k.r.bookcustomer.Resilience4jUnitTest  : meet error RateLimiter 'name1' does not permit further calls
10. 2022-12-28T15:39:52.075+08:00 ERROR 35236 --- [onPool-worker-4] c.k.r.bookcustomer.Resilience4jUnitTest  : meet error RateLimiter 'name1' does not permit further calls

复制代码

限速器这个功能只能限制在整体性能上面，如果要限制某个用户，只能某段时间访问多少次，它就做不到了。
Relilience4j 里面还提供了Retry,TimeLimiter,Cache. 感觉不是很有必要的功能， Retry在spring里面有相应的功能了，没有必要专门为了使用它而多加个包。 TimeLimiter，Cache 我感觉不是很受重视的功能，连例子文档都懒得提供，可见意义不大。

免责声明：如果侵犯了您的权益，请联系站长，我们会及时删除侵权内容，谢谢合作！