Spring Boot微服务中Resilience4j熔断与重试共存时重试失效问题

问题场景 #

在基于Spring Boot的微服务项目中，两个微服务通过API网关交互，尝试用Resilience4j同时实现熔断和重试机制，但发现添加@CircuitBreaker注解后重试逻辑完全不执行，移除该注解后重试功能恢复正常。

服务层核心代码 #

@Override
@CircuitBreaker(name="${spring.application.name}", fallbackMethod = "getDefaultDepartment")
@Retry(name="${spring.application.name}", fallbackMethod = "getDefaultDepartment")
public ResponseDto getEmployee(Long id) {
    LOGGER.info("inside getEmployee by ID method");
    Employee employee = employeeRepository.findById(id).orElseThrow(() -> new ResourceNotFoundException("Employee", "id", id));
    
    // OpenFeign调用其他微服务
    DepartmentDto departmentDto = apiClient.getDepartment(employee.getDepartmentCode());

    return new ResponseDto(modelMapper.map(employee, EmployeeDto.class), departmentDto);
}

配置文件内容 #

#------------------------------  Circuit breaker properties--------------------------------------
resilience4j.circuitbreaker.instances.EMPLOYEE-SERVICE.registerHealthIndicator=true
resilience4j.circuitbreaker.instances.EMPLOYEE-SERVICE.failureRateThreshold=50
resilience4j.circuitbreaker.instances.EMPLOYEE-SERVICE.minimumNumberOfCalls=5
resilience4j.circuitbreaker.instances.EMPLOYEE-SERVICE.automaticTransitionFromOpenToHalfOpenEnabled=true
resilience4j.circuitbreaker.instances.EMPLOYEE-SERVICE.waitDurationInOpenState=5s
resilience4j.circuitbreaker.instances.EMPLOYEE-SERVICE.permittedNumberOfCallsInHalfOpenState=3
resilience4j.circuitbreaker.instances.EMPLOYEE-SERVICE.slidingWindowSize=10
resilience4j.circuitbreaker.instances.EMPLOYEE-SERVICE.slidingWindowType=COUNT_BASED


#------------------------------  Retry Configuration  --------------------------------------
resilience4j.retry.instances.EMPLOYEE-SERVICE.registerHealthIndicator=true
resilience4j.retry.instances.EMPLOYEE-SERVICE.maxAttempts=10
resilience4j.retry.instances.EMPLOYEE-SERVICE.waitDuration=1s

原因分析 #

Resilience4j的注解基于Spring AOP实现，默认情况下@CircuitBreaker的AOP优先级高于@Retry。这意味着熔断的代理逻辑会包裹重试的代理逻辑：当调用失败时，熔断会第一时间捕获异常并触发fallback，根本不会进入重试的逻辑分支。

解决方案 #

1. 调整注解顺序（最简单有效） #

将@Retry注解放在@CircuitBreaker上方，让重试的代理包裹熔断的代理。这样调用失败时会先执行重试逻辑，重试耗尽后再触发熔断：

@Override
@Retry(name="${spring.application.name}", fallbackMethod = "getDefaultDepartment")
@CircuitBreaker(name="${spring.application.name}", fallbackMethod = "getDefaultDepartment")
public ResponseDto getEmployee(Long id) {
    // 原有逻辑不变
}

2. 自定义AOP优先级（灵活控制） #

如果需要更精细的优先级控制，可以通过配置类指定Retry和CircuitBreaker的AOP顺序，确保Retry的优先级更高：

import io.github.resilience4j.circuitbreaker.CircuitBreakerAspect;
import io.github.resilience4j.retry.RetryAspect;
import org.springframework.context.annotation.Bean;
import org.springframework.context.annotation.Configuration;
import org.springframework.core.Ordered;

@Configuration
public class Resilience4jConfig {

    @Bean
    public RetryAspect retryAspect() {
        RetryAspect aspect = new RetryAspect();
        aspect.setOrder(Ordered.HIGHEST_PRECEDENCE); // 最高优先级
        return aspect;
    }

    @Bean
    public CircuitBreakerAspect circuitBreakerAspect() {
        CircuitBreakerAspect aspect = new CircuitBreakerAspect();
        aspect.setOrder(Ordered.HIGHEST_PRECEDENCE + 1); // 优先级低于Retry
        return aspect;
    }
}

3. 确保重试捕获目标异常 #

默认情况下，Resilience4j Retry只会重试RuntimeException类型的异常。如果Feign调用抛出的FeignException或自定义异常（如ResourceNotFoundException）需要被重试，需在配置中明确指定：

resilience4j.retry.instances.EMPLOYEE-SERVICE.retry-exceptions=com.example.ResourceNotFoundException,feign.FeignException

4. 统一fallback逻辑注意事项 #

两个注解共用同一个fallback方法时，需确保方法签名兼容：fallback方法需要接收原方法的参数+异常对象，示例如下：

public ResponseDto getDefaultDepartment(Long id, Throwable throwable) {
    // 降级逻辑实现
    EmployeeDto employeeDto = new EmployeeDto();
    DepartmentDto departmentDto = new DepartmentDto();
    departmentDto.setDepartmentName("Default Dept");
    departmentDto.setDepartmentCode("DEFAULT");
    return new ResponseDto(employeeDto, departmentDto);
}

内容的提问来源于stack exchange，提问作者Somesh Vemula