Hystrix停更之后,Netflix官方推荐移步至resilience4j,它是一个轻量、易用、可组装的高可用框架,支持熔断、高频控制、隔离、限流、限时、重试等多种高可用机制。
与Hystrix相比,它有以下一些主要的区别:
下面依次介绍各类高可用机制的使用方式【选译官方使用文档】。
CircuitBreakerRegistry负责创建和管理熔断器实例CircuitBreaker,它是线程安全的,提供原子性操作。
CircuitBreakerRegistry circuitBreakerRegistry = CircuitBreakerRegistry.ofDefaults();
该方式使用默认的全局配置CircuitBreakerConfig创建熔断器实例,你也可以选择使用定制化的配置,可选项有:
// 创建定制化熔断器配置
CircuitBreakerConfig circuitBreakerConfig = CircuitBreakerConfig.custom()
.failureRateThreshold(50)
.waitDurationInOpenState(Duration.ofMillis(1000))
.ringBufferSizeInHalfOpenState(2)
.ringBufferSizeInClosedState(2)
.build();
// 使用定制化配置创建熔断器注册中心
CircuitBreakerRegistry circuitBreakerRegistry = CircuitBreakerRegistry.of(circuitBreakerConfig);
// 从注册中心获取使用默认配置的熔断器
CircuitBreaker circuitBreaker2 = circuitBreakerRegistry.circuitBreaker("otherName");
// 从注册中心获取使用定制化配置的熔断器
CircuitBreaker circuitBreaker = circuitBreakerRegistry.circuitBreaker("uniqueName", circuitBreakerConfig);
你也可以选择不经过注册中心,直接创建熔断器实例:
CircuitBreaker defaultCircuitBreaker = CircuitBreaker.ofDefaults("testName");
CircuitBreaker customCircuitBreaker = CircuitBreaker.of("testName", circuitBreakerConfig);
熔断器采用装饰器模式,你可以使用CircuitBreaker.decorateCheckedSupplier() / CircuitBreaker.decorateCheckedRunnable() / CircuitBreaker.decorateCheckedFunction() 装饰 Supplier / Runnable / Function / CheckedRunnable / CheckedFunction,然后使用Vavr的Try.of(…) / Try.run(…) 调用被装饰的函数,也可以使用map / flatMap / filter / recover / andThen链接更多的函数。函数链只有在熔断器处于关闭或半开状态时才可以被调用。
// 创建熔断器
CircuitBreaker circuitBreaker = CircuitBreaker.ofDefaults("testName");
// 用熔断器包装函数
CheckedFunction0 decoratedSupplier = CircuitBreaker
.decorateCheckedSupplier(circuitBreaker, () -> "This can be any method which returns: 'Hello");
// 链接其它的函数
Try result = Try.of(decoratedSupplier)
.map(value -> value + " world'");
// 如果函数链中的所有函数均调用成功,最终结果为Success
assertThat(result.isSuccess()).isTrue();
assertThat(result.get()).isEqualTo("This can be any method which returns: 'Hello world'");
函数链中可以包含被不同熔断器包装的多个函数:
// 两个熔断器
CircuitBreaker circuitBreaker = CircuitBreaker.ofDefaults("testName");
CircuitBreaker anotherCircuitBreaker = CircuitBreaker.ofDefaults("anotherTestName");
// 用两个熔断器分别包装Supplier 和 Function
CheckedFunction0 decoratedSupplier = CircuitBreaker
.decorateCheckedSupplier(circuitBreaker, () -> "Hello");
CheckedFunction1 decoratedFunction = CircuitBreaker
.decorateCheckedFunction(anotherCircuitBreaker, (input) -> input + " world");
// 链接函数
Try result = Try.of(decoratedSupplier)
.mapTry(decoratedFunction::apply);
assertThat(result.isSuccess()).isTrue();
assertThat(result.get()).isEqualTo("Hello world");
下面我们模拟熔断器被触发的情况,在熔断器打开的状态,Try.of 返回 Failure
// 创建一个环状缓冲区大小为2的熔断器
CircuitBreakerConfig circuitBreakerConfig = CircuitBreakerConfig.custom()
.ringBufferSizeInClosedState(2)
.waitDurationInOpenState(Duration.ofMillis(1000))
.build();
CircuitBreaker circuitBreaker = CircuitBreaker.of("testName", circuitBreakerConfig);
// 模拟一次失败调用
circuitBreaker.onError(0, new RuntimeException());
// 没有触发熔断,熔断器仍处于关闭状态
assertThat(circuitBreaker.getState()).isEqualTo(CircuitBreaker.State.CLOSED);
// 模拟第二次失败调用
circuitBreaker.onError(0, new RuntimeException());
// 失败率超过百分之五十,熔断器被触发
assertThat(circuitBreaker.getState()).isEqualTo(CircuitBreaker.State.OPEN);
// 由于熔断器处于打开状态,调用失败
Try result = Try.of(CircuitBreaker.decorateCheckedSupplier(circuitBreaker, () -> "Hello"))
.map(value -> value + " world");
assertThat(result.isFailure()).isTrue();
assertThat(result.failed().get()).isInstanceOf(CircuitBreakerOpenException.class);
熔断器支持RxJava/Reactor,下面是简单的示例:
CircuitBreaker circuitBreaker = CircuitBreaker.ofDefaults("backendName");
Observable.fromCallable(backendService::doSomething)
.lift(CircuitBreakerOperator.of(circuitBreaker))
CircuitBreaker circuitBreaker = CircuitBreaker.ofDefaults("backendName");
Mono.fromCallable(backendService::doSomething)
.transform(CircuitBreakerOperator.of(circuitBreaker))
熔断器支持重置,重置之后所有状态数据清空,恢复初始状态。
circuitBreaker.reset();
你可以为熔断函数设置降级措施,链接Try.recover(),当Try.of() 返回 Failure
// 创建熔断器
CircuitBreaker circuitBreaker = CircuitBreaker.ofDefaults("testName");
// 模拟失败调用,并链接降级函数
CheckedFunction0 checkedSupplier = CircuitBreaker.decorateCheckedSupplier(circuitBreaker, () -> {
throw new RuntimeException("BAM!");
});
Try result = Try.of(checkedSupplier)
.recover(throwable -> "Hello Recovery");
// 降级函数被调用,最终调用结果为成功
assertThat(result.isSuccess()).isTrue();
assertThat(result.get()).isEqualTo("Hello Recovery");
默认所有异常都被认定为失败事件,你可以定制Predicate的test检查,实现选择性记录,只有该函数返回为true时异常才被认定为失败事件。下例展示了如何忽略除WebServiceException外的所有异常:
CircuitBreakerConfig circuitBreakerConfig = CircuitBreakerConfig.custom()
.ringBufferSizeInClosedState(2)
.ringBufferSizeInHalfOpenState(2)
.waitDurationInOpenState(Duration.ofMillis(1000))
.recordFailure(throwable -> Match(throwable).of(
Case($(instanceOf(WebServiceException.class)), true),
Case($(), false)))
.build();
熔断器事件CircuitBreakerEvent包含状态转换、重置、成功、失败异常、忽略异常事件,所有事件包含发生时间、处理时长的信息。你可以使用如下方式注册事件监听器:
// 分类监听事件
circuitBreaker.getEventPublisher()
.onSuccess(event -> logger.info(...))
.onError(event -> logger.info(...))
.onIgnoredError(event -> logger.info(...))
.onReset(event -> logger.info(...))
.onStateTransition(event -> logger.info(...));
// 监听所有事件
circuitBreaker.getEventPublisher()
.onEvent(event -> logger.info(...));
你也可以使用CircularEventConsumer将事件存储在缓存中:
CircularEventConsumer ringBuffer = new CircularEventConsumer<>(10);
circuitBreaker.getEventPublisher().onEvent(ringBuffer);
List bufferedEvents = ringBuffer.getBufferedEvents()
你还可以将EventPublisher转换为RxJava/Reactor的事件流,该方法的优势在于,可以进一步指定调度器进行异步化处理:
RxJava2Adapter.toFlowable(circuitBreaker.getEventPublisher())
.filter(event -> event.getEventType() == Type.ERROR)
.cast(CircuitBreakerOnErrorEvent.class)
.subscribe(event -> logger.info(...))
CircuitBreaker.Metrics metrics = circuitBreaker.getMetrics();
float failureRate = metrics.getFailureRate();
int bufferedCalls = metrics.getNumberOfBufferedCalls();
int failedCalls = metrics.getNumberOfFailedCalls();
高频控制的配置方式与熔断类似,有对应的RateLimiterRegistry 和 RateLimiterConfig,自定义配置的可选项有:
// 创建高频控制配置,控制频率为1QPS
RateLimiterConfig config = RateLimiterConfig.custom()
.limitForPeriod(1)
.limitRefreshPeriod(Duration.ofMillis(1000))
.timeoutDuration(Duration.ofMillis(500))
.build();
// 创建高频控制注册中心
RateLimiterRegistry rateLimiterRegistry = RateLimiterRegistry.of(config);
// 从注册中心创建高频控制器实例
RateLimiter rateLimiterWithDefaultConfig = rateLimiterRegistry.rateLimiter("backend");
RateLimiter rateLimiterWithCustomConfig = rateLimiterRegistry.rateLimiter("backend#2", config);
// 直接创建高频控制器实例
RateLimiter rateLimiter = RateLimiter.of("NASDAQ :-)", config);
// 使用上面定义的高频控制器装饰函数调用
CheckedRunnable restrictedCall = RateLimiter
.decorateCheckedRunnable(rateLimiter, () -> System.out.println("Do something"));
// 第一次调用成功,第二次调用被高频限制
Try.run(restrictedCall)
.andThenTry(restrictedCall)
.onFailure(throwable -> System.out.println("Wait before call it again :)"));
你可以在运行时动态修改高频控制器配置,但新的冷却时间不会影响当前处于冷却状态的线程,新的阈值也不会影响处于当前一轮控制的线程:
// 在下一轮控制中,阈值变更为100
rateLimiter.changeLimitForPeriod(100);
与熔断类似,高频控制也支持RxJava/Reactor.
RateLimiter rateLimiter = RateLimiter.ofDefaults("backendName");
Observable.fromCallable(backendService::doSomething)
.lift(RateLimiterOperator.of(rateLimiter));
RateLimiter rateLimiter = RateLimiter.ofDefaults("backendName");
Mono.fromCallable(backendService::doSomething)
.transform(RateLimiterOperator.of(rateLimiter));
高频控制器事件RateLimiterEvent包含允许执行和拒绝执行事件,所有事件包含发生时间、相关高频控制器名称的信息。
rateLimiter.getEventPublisher()
.onSuccess(event -> logger.info(...))
.onFailure(event -> logger.info(...));
将EventPublisher转换为RxJava/Reactor的事件流:
RxJava2Adapter.toFlowable(rateLimiter.getEventPublisher())
.filter(event -> event.getEventType() == FAILED_ACQUIRE)
.subscribe(event -> logger.info(...))
RateLimiter limit;
RateLimiter.Metrics metrics = limit.getMetrics();
int numberOfThreadsWaitingForPermission = metrics.getNumberOfWaitingThreads();
int availablePermissions = metrics.getAvailablePermissions();
AtomicRateLimiter atomicLimiter;
long nanosToWaitForPermission = atomicLimiter.getNanosToWait();
Bulkhead意指船舶中的隔舱板,它将船体分割为多个船舱,在船部分受损时可避免沉船。框架中的Bulkhead通过信号量的方式隔离不同种类的调用,并进行流控,这样可以避免某类调用异常危及系统整体。(注:不同于Hystrix,该框架不提供基于线程池的隔离)
Bulkhead的配置方式与熔断类似,有对应的BulkheadRegistry 和 BulkheadConfig,自定义配置的可选项有:
// 创建自定义的Bulkhead配置
BulkheadConfig config = BulkheadConfig.custom()
.maxConcurrentCalls(150)
.maxWaitTime(100)
.build();
// 创建Bulkhead注册中心
BulkheadRegistry registry = BulkheadRegistry.of(config);
// 从注册中心获取默认配置的Bulkhead
Bulkhead bulkhead1 = registry.bulkhead("foo");
// 从注册中心获取自定义配置的Bulkhead
Bulkhead bulkhead2 = registry.bulkhead("bar", config);
你也可以不经过注册中心,直接创建Bulkhead:
Bulkhead bulkhead1 = Bulkhead.ofDefaults("foo");
Bulkhead bulkhead2 = Bulkhead.of(
"bar",
BulkheadConfig.custom()
.maxConcurrentCalls(50)
.build()
);
与熔断类似,不再赘述,下面是代码示例:
// 创建Bulkhead实例
Bulkhead bulkhead = Bulkhead.of("testName", config);
// 用Bulkhead包装函数调用
CheckedFunction0 decoratedSupplier = Bulkhead.decorateCheckedSupplier(bulkhead, () -> "This can be any method which returns: 'Hello");
// 链接其它函数
Try result = Try.of(decoratedSupplier)
.map(value -> value + " world'");
assertThat(result.isSuccess()).isTrue();
assertThat(result.get()).isEqualTo("This can be any method which returns: 'Hello world'");
assertThat(bulkhead.getMetrics().getAvailableConcurrentCalls()).isEqualTo(1);
函数链中可以包含被不同Bulkhead或熔断器包装的多个函数:
// 两个Bulkhead
Bulkhead bulkhead = Bulkhead.of("test", config);
Bulkhead anotherBulkhead = Bulkhead.of("testAnother", config);
// 用两个Bulkhead分别包装Supplier 和 Function
CheckedFunction0 decoratedSupplier
= Bulkhead.decorateCheckedSupplier(bulkhead, () -> "Hello");
CheckedFunction1 decoratedFunction
= Bulkhead.decorateCheckedFunction(anotherBulkhead, (input) -> input + " world");
// 链接函数
Try result = Try.of(decoratedSupplier)
.mapTry(decoratedFunction::apply);
assertThat(result.isSuccess()).isTrue();
assertThat(result.get()).isEqualTo("Hello world");
assertThat(bulkhead.getMetrics().getAvailableConcurrentCalls()).isEqualTo(1);
assertThat(anotherBulkhead.getMetrics().getAvailableConcurrentCalls()).isEqualTo(1);
下面我们模拟饱和态Bulkhead的行为:
// 创建并行度为2的Bulkhead
BulkheadConfig config = BulkheadConfig.custom().maxConcurrentCalls(2).build();
Bulkhead bulkhead = Bulkhead.of("test", config);
// 该方法会进入Bulkhead
bulkhead.isCallPermitted();
bulkhead.isCallPermitted();
// 经过上面两次调用,Bulkhead已饱和
CheckedRunnable checkedRunnable = Bulkhead.decorateCheckedRunnable(bulkhead, () -> {throw new RuntimeException("BAM!");});
Try result = Try.run(checkedRunnable);
assertThat(result.isFailure()).isTrue();
assertThat(result.failed().get()).isInstanceOf(BulkheadFullException.class);
你可以在运行时动态修改Bulkhead配置,但新的最大阻塞时间不会影响当前正在等待的线程。
与熔断类似,Bulkhead也支持RxJava/Reactor.
Bulkhead bulkhead = Bulkhead.ofDefaults("backendName");
Observable.fromCallable(backendService::doSomething)
.lift(BulkheadOperator.of(bulkhead));
Bulkhead bulkhead = Bulkhead.ofDefaults("backendName");
Mono.fromCallable(backendService::doSomething)
.transform(BulkheadOperator.of(bulkhead));
BulkHeadEvent包含允许执行、拒绝执行、执行结束事件,可以用如下方式监听:
bulkhead.getEventPublisher()
.onCallPermitted(event -> logger.info(...))
.onCallRejected(event -> logger.info(...))
.onCallFinished(event -> logger.info(...));
Bulkhead.Metrics metrics = bulkhead.getMetrics();
in remainingBulkheadDepth = metrics.getAvailableConcurrentCalls()
限时器与Future配合使用,限时器将Future Supplier转换为Callable,它将尝试在给定时间内获取Future的值,如果超时未获取到,Future将会被取消:
// 创建限时器配置,最大执行时间为60s,超时将取消Future
TimeLimiterConfig config = TimeLimiterConfig.custom()
.timeoutDuration(Duration.ofSeconds(60))
.cancelRunningFuture(true)
.build();
// 从该配置创建限时器
TimeLimiter timeLimiter = TimeLimiter.of(config);
ExecutorService executorService = Executors.newSingleThreadExecutor();
// 将待执行任务提交到线程池,获取Future Supplier
Supplier> futureSupplier = () -> executorService.submit(backendService::doSomething)
// 使用限时器包装Future Supplier
Callable restrictedCall = TimeLimiter
.decorateFutureSupplier(timeLimiter, futureSupplier);
// 若任务执行超时,onFailure会被触发
Try.of(restrictedCall.call)
.onFailure(throwable -> LOG.info("A timeout possibly occurred."));
你可以将限时器与熔断器配合使用,在超时次数过多时触发熔断:
Callable restrictedCall = TimeLimiter
.decorateFutureSupplier(timeLimiter, futureSupplier);
Callable chainedCallable = CircuitBreaker.decorateCallable(circuitBreaker, restrictedCall);
Try.of(chainedCallable::call)
.onFailure(throwable -> LOG.info("We might have timed out or the circuit breaker has opened."));
可配置选项有:
// 最多重试2次
// 重试间隔为100ms
// 当执行出现WebServiceException时触发重试
RetryConfig config = RetryConfig.custom()
.maxAttempts(2)
.waitDuration(Duration.ofMillis(100))
.retryOnException(throwable -> API.Match(throwable).of(
API.Case($(Predicates.instanceOf(WebServiceException.class)), true),
API.Case($(), false)))
.build();
// 从重试配置创建重试器
Retry retry = Retry.of("id", config);
// 使用重试器包装函数调用
CheckedFunction0 retryableSupplier = Retry.decorateCheckedSupplier(retry, () -> {throw new WebServiceException("BAM!");});
// 调用将会自动重试
Try result = Try.of(retryableSupplier).recover((throwable) -> "Hello world from recovery function");
assertThat(result.get()).isEqualTo("Hello world from recovery function");
io.github.resilience4j
resilience4j-circuitbreaker
0.13.1
io.github.resilience4j
resilience4j-ratelimiter
0.13.1
io.github.resilience4j
resilience4j-retry
0.13.1
io.github.resilience4j
resilience4j-bulkhead
0.13.1
io.github.resilience4j
resilience4j-cache
0.13.1
io.github.resilience4j
resilience4j-timelimiter
0.13.1