MockRestServiceServer未及时更新运行状况终结点的restTemplate

Question

我最近将SpringBoot项目从1.1升级到1.4，突然，“/health”endpoint的测试开始失败

@SpringBootTest
class HealthTest extends Specification {

  @Autowired
  RestTemplate thirdPartyRestTemplate

  def 'should set health status based on third party service'() {
    given:
    MockRestServiceServer thirdPartyServerMock = MockRestServiceServer.createServer(thirdPartyRestTemplate)

    thirdPartyServerMock
      .expect(MockRestRequestMatchers.requestTo('/status'))
      .andExpect(MockRestRequestMatchers.method(HttpMethod.GET))
      .andRespond(MockRestResponseCreators.withStatus(thirdPartyServerResponse).body('{}'))

    when:
    def response = RestAssured.given().get('/health')
    println(response.statusCode)
    Map health = response.as(Map)

    then:
    thirdPartyServerMock.verify()
    health == expectedHealth

    where:
    thirdPartyServerResponse       | expectedHealth
    HttpStatus.OK                  | [status: 'UP', thirdPartyServer: 'UP']
    HttpStatus.SERVICE_UNAVAILABLE | [status: 'DOWN', thirdPartyServer: 'DOWN']
  }

}

发生的事情是：第一个测试总是通过，而第二个测试总是失败。输出为200。如果顺序颠倒，也会发生同样的情况，所以

where:
    thirdPartyServerResponse       | expectedHealth
    HttpStatus.SERVICE_UNAVAILABLE | [status: 'DOWN', thirdPartyServer: 'DOWN']
    HttpStatus.OK                  | [status: 'UP', thirdPartyServer: 'UP']

这一次，503失败了。如果我添加线程。在像这样的实际Rest电话之前先睡一觉

when:
    Thread.sleep(1000)
    def response = RestAssured.given().get('/health')

那它每次都会过去！因此，Spring似乎在MockRestServiceServer中引入了一些更改，并且需要一些时间来配置一个模拟（可能现在在单独的线程中执行）。

有人有类似的问题吗？如何绕过此线程。睡眠修复和这种行为的解释是什么？

Answer 1

事实证明，Spring为健康endpoint引入了缓存机制，默认情况下，它将结果缓存1秒。从文档中：

# Time to live for cached result, in milliseconds.    
endpoints.health.time-to-live=1000 

一旦我将配置更改为：

endpoints:
  health:
    time-to-live: 0

考试又开始通过了。