spring-boot - 如何正确配置 Kubernetes 探针计时(针对 Spring Boot 应用程序)

Question

我们有一个简单的 Spring Boot Web 应用程序，启动时间不到 30 秒。所以我按如下方式配置探针:

    readinessProbe:
      httpGet:
        path: /actuator/health/readiness
        port: 8080
      initialDelaySeconds: 30
      periodSeconds: 1
    livenessProbe:
      httpGet:
        path: /actuator/health/liveness
        port: 8080
      initialDelaySeconds: 30
      periodSeconds: 1

我的理解是，就绪探测会等待 30 秒，然后就会成功(如果应用程序已启动)。而且， active 探测会延迟 30 秒(从部署开始)启动，并且几乎会在就绪性探测成功的同时成功(如果应用程序已准备就绪)。但我在日志中看到的是，就绪探针等待 30 秒然后成功，但之后又等待 30 秒，然后旧 Pod 被关闭:

develop/demo-57c8984866-6v5sl[demo]: {"ts":"2020-10-08T17:33:07.378Z","logger":"org.springframework.boot.web.embedded.tomcat.TomcatWebServer","level":"INFO","thread":"main","message":"Tomcat started on port(s): 8080 (http) with context path ''"}
develop/demo-57c8984866-6v5sl[demo]: {"ts":"2020-10-08T17:33:07.387Z","logger":"org.springframework.data.repository.config.DeferredRepositoryInitializationListener","level":"INFO","thread":"main","message":"Triggering deferred initialization of Spring Data repositories…"}
develop/demo-57c8984866-6v5sl[demo]: {"ts":"2020-10-08T17:33:09.441Z","logger":"org.springframework.data.repository.config.DeferredRepositoryInitializationListener","level":"INFO","thread":"main","message":"Spring Data repositories initialized!"}
develop/demo-57c8984866-6v5sl[demo]: {"ts":"2020-10-08T17:33:09.469Z","logger":"com.example.Application","level":"INFO","thread":"main","message":"Started Application in 23.918 seconds (JVM running for 25.343)"}
develop/demo-57c8984866-6v5sl[demo]: {"ts":"2020-10-08T17:33:14.251Z","logger":"org.apache.catalina.core.ContainerBase.[Tomcat].[localhost].[/]","level":"INFO","thread":"http-nio-8080-exec-1","message":"Initializing Spring DispatcherServlet 'dispatcherServlet'"}
develop/demo-57c8984866-6v5sl[demo]: {"ts":"2020-10-08T17:33:14.258Z","logger":"org.springframework.web.servlet.DispatcherServlet","level":"INFO","thread":"http-nio-8080-exec-1","message":"Initializing Servlet 'dispatcherServlet'"}
develop/demo-57c8984866-6v5sl[demo]: {"ts":"2020-10-08T17:33:14.292Z","logger":"org.springframework.web.servlet.DispatcherServlet","level":"INFO","thread":"http-nio-8080-exec-1","message":"Completed initialization in 30 ms"}
develop/demo-79cc9bc757-xlg6z[demo]: 2020-10-08T17:33:44.590172 Shutting down...
develop/demo-79cc9bc757-xlg6z[demo]: {"ts":"2020-10-08T17:33:44.658Z","logger":"org.springframework.orm.jpa.LocalContainerEntityManagerFactoryBean","level":"INFO","thread":"SpringContextShutdownHook","message":"Closing JPA EntityManagerFactory for persistence unit 'default'"}
develop/demo-79cc9bc757-xlg6z[demo]: {"ts":"2020-10-08T17:33:44.664Z","logger":"org.springframework.scheduling.concurrent.ThreadPoolTaskExecutor","level":"INFO","thread":"SpringContextShutdownHook","message":"Shutting down ExecutorService 'applicationTaskExecutor'"}
develop/demo-79cc9bc757-xlg6z[demo]: {"ts":"2020-10-08T17:33:44.667Z","logger":"com.zaxxer.hikari.HikariDataSource","level":"INFO","thread":"SpringContextShutdownHook","message":"HikariPool-1 - Shutdown initiated..."}
develop/demo-79cc9bc757-xlg6z[demo]: {"ts":"2020-10-08T17:33:44.680Z","logger":"com.zaxxer.hikari.HikariDataSource","level":"INFO","thread":"SpringContextShutdownHook","message":"HikariPool-1 - Shutdown completed."}

第二个30秒的等待时间是什么原因(参见第一个“正在关闭...”)？根据记录，不存在 SIGTERM 问题(应用程序正确响应 SIGTERM)。

更多信息
如果我对这些探测器的工作方式有误，请纠正我:容器启动，此时准备初始延迟的计时器也启动。 25 秒后，应用程序准备就绪。再过 5 秒，就绪探针开始命中应用程序/容器并且成功，因此应用程序已准备就绪(对吗？)。此时，我希望 k8s 向旧 pod 发送 sigterm，要求其关闭。但如日志所示，新容器准备就绪后，旧 Pod 仍运行了 30 秒以上。
也许，重新表述这个问题会有所帮助。我想在应用程序准备好后立即关闭旧的 Pod。该应用程序在 30 秒内即可准备就绪。换句话说，我需要整个部署只需要 30 秒(考虑到启动时间不到 30 秒，关闭时间不到 1 秒)。为什么上面的配置没有做到这一点？ 谢谢。

Answer 1

好吧，事实证明，我们用来管理 k8s list 的工具 ( kubecfg ) 和库的 minReadySeconds 属性具有不同的默认值(30 秒)。我改变了这一点，一切都按预期进行。