从应用开发角度认识K8S

{"type":"doc","content":[{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"云原生应用"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们正经历从单体应用转向分布式微服务架构应用的技术趋势。分布式微服务架构作为越来越多的软件开发设计模式，以领域设计模型来指导业务需求的抽象与封装。对业务的实体抽象还是边界划分，会以微服务架构作为落地点，形成微服务集群。并实施运行在云原生编排平台。"}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/01/013b22edc66739cb3f42827eb6c55f6c.png","alt":null,"title":"云原生应用结构 from Kubernetes-Patterns","style":[{"key":"width","value":"50%"},{"key":"bordertype","value":"none"}],"href":"","fromPaste":false,"pastePass":false}},{"type":"bulletedlist","content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"云原生应用的基石是干净整洁，业务逻辑相对单一，并与其他领域对象独立的代码实现。这一阶段保证业务质量的主要是编程基本功，以及高覆盖率的自动化测试能力。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"领域设计驱动是近年微服务技术热潮下的主流设计模式，主要解决的问题是如何拆解一个复杂的业务场景需求到多个微服务单元。领域设计驱动是微服务架构的设计模式，微服务架构是基于领域设计模式的实现方式。一个微服务可以对应一个领域对象，也可以是一个领域服务。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"分布式微服务架构实现的云原生应用具有高可用，弹性伸缩，容忍失败以及健康自省等特点。它使得我们处理日益增长的业务需求的能力从开发编程的复杂性逐渐转移到了资源整合，操作与管理的复杂性上。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"微服务是单一的，运行在一个单进程中的简单应用。容器技术恰好能够提供这种隔离封装，将一个简单的微服务以Dockfile模版方式标准化，可无差别得运行在分布式集群的任意资源节点。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"K8S作为目前最流行的云原生平台架构，对于一组微服务的交互，持续化数据的存储，或者实施多个具有依赖关系的微服务运行，以及容量规划等问题，能够提供一套自动化的系统性解决方案。"}]}]}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"用OOP方式解读K8S"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"对于应用开发者，面向对象模式想必了然于胸。OOP设计了一套对一个逻辑对象的生命周期管理的方法论，类比OOP思路，笔者接下来详细介绍一些K8S核心资源对象以及应用方式。"}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/a1/a1495432cb615caf9d3a76fb8448aac9.png","alt":null,"title":"OOP vs K8S from Kubernetes-Patterns","style":[{"key":"width","value":"100%"},{"key":"bordertype","value":"none"}],"href":"","fromPaste":false,"pastePass":false}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"构建/部署保持隔离性Pod/Deployment"}]},{"type":"bulletedlist","content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"Image"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"容器镜像类比OOP的类，定义了一个模块的全部属性与功能，提供了唯一暴露在外的API调用方式以及参数集合，对应着一个独立完整的发布周期，就像容器的设计蓝图。这种静态定义方式，可以定义并初始化容器进行，使其在任意环境任意时刻行为完全一致。一个容器镜像对应着一个微服务，属于开发团队的产物。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"Container"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"容器类比OOP的对象，是容器镜像的运行态。一个容器是一个容器镜像的运行进程，而一个容器镜像则可以在任意时刻任意环境下创建任何数量容器。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"Pod"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Java应用开发者都知道基于Springboot-MVC框架的Java应用部署时只需提供一个Jar包，Jar包内部源码被编译后不可再改变。Pod是云原生编排平台的资源调度部署的最小单元。Pod和容器的关系类似Java的Jar包和对象，应用开发者交付的容器镜像通过Pod在K8S集群上部署并调度，容器则是Pod的内部资源对象，K8S无法感知也无法干预。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"K8S设计Pod为部署调度的最小单元，是因为Pod实现了内部一组容器在存储空间，网络空间及进程空间可共享该Pod资源，类似一个虚拟机上同时运行多个进程。容器间通信类似单节点进程间通信。Pod 的设计，就是要让它里面的容器尽可能多地共享 Linux Namespace，仅保留必要的隔离和限制能力。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Pod类似OOP的Module，即逻辑紧密的对象集合通常属于一个独立模块。Pod的定义示例如下："}]}]}]},{"type":"codeblock","attrs":{"lang":"text"},"content":[{"type":"text","text":"apiVersion: v1\nkind: Pod\nmetadata:\n name: index-helm-57677c549-lgww5\n namespace: bss-dev\nspec:\n containers:\n - command:\n - java\n - '-jar'\n - /home/demo/app.jar\n env:\n - name: aliyun_logs_release_tags\n value: revision=3f44253.20201030-1039\n image: 'registry-vpc.cn-shanghai-finance-1.aliyuncs.com/XXX/XXX.XXX:latest'\n imagePullPolicy: Always"}]},{"type":"bulletedlist","content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Pod->spec下包含了一个或多个容器模版定义。PodYAML提供了很多属性，有兴趣深入的读者可以参考学习"},{"type":"link","attrs":{"href":"https://github.com/kubernetes/kubernetes/blob/master/staging/src/k8s.io/api/core/v1/types.go","title":""},"content":[{"type":"text","text":"https://github.com/kubernetes/kubernetes/blob/master/staging/src/k8s.io/api/core/v1/types.go"}]}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Pod->spec有一个很特殊的容器模版定义，关键词为initContainer，顾名思义是做初始化的容器。初始化容器必须先于应用容器启动执行完毕，并且只能执行成功。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Pod->spec除了定义了容器模版，还定义容器间共享的存储资源Volume挂载方式，和影响Pod资源调度的节点选择器标签，亲和性以及容忍性等。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"NameSpace"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"命名空间是一个组的概念，为集群资源提供了逻辑划分的能力。这种使用方式类似OOP的Package。当项目变大，时长有同名的类或者对象，此时为了区分，我们会并以packge为路径前缀，定义和引用相应对象。K8S集群中常常运行着数百个应用服务，也会出现同名资源的情况。命名空间可以在K8S上实现对一组资源对象隔离与权限管理。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"命名空间最常用的场景是在一个K8S集群区分开发环境和测试环境。命名空间也可以提供多租户运行环境，或者为了应用运行具有隔离性，为某个应用部署命名一个单独的命名空间。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"命名空间虽然提供了资源范围逻辑划分的能力，但是并没有真正隔离一个集群内部Pod之间的通信，即属于同一集群内的多个命名空间下的Pod仍能在集群内互相通信。如果需要做到命名空间之间的完全隔离，可以采用NetworkPolicy实现。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"Deployment"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在K8S平台运行的应用服务版本升级就是创建一个新版本Pod，并销毁旧版本Pod的过程，由Deployment资源对象定义。微服务架构应用的数量成百上千，如果手动部署，则会引入人为错误并且部署操作很容易成为整个系统的瓶颈。K8S将部署Pod全部操作定义在Deployment资源对象，由平台自动化地部署Pod。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Deployment资源对象除了要定义部署Pod是什么样的以外，还会定义预期部署状态。比如，预期部署Pod数量，或在哪个节点上部署。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"部署一个新版本要么就生成额外的带有新版本Pod，待健康检查等确认新Pod可以提供服务，再将原有的带有旧版本Pod销毁，最终达到预期部署状态；要么就先销毁原有旧版本Pod，再生成新版本Pod，直到达到预期部署状态。第一种方式为Rolling Update，好处是部署时没有任何downtime，坏处是会存在多个版本同时提供服务，导致服务状态不一致。第二种方式为Recreate，好处是在任意时刻都不会存在多个版本的应用服务，坏处则是部署时存在downtime。K8S默认的部署策略为Rolling Update。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":1,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Deployment通过ReplicaSet控制器创建和管理Pod，确保Pod能如预期运行成功，并且满足Deployment的部署定义，比如开启几个副本，满足部署策略等。ReplicaSet解耦了部署与Pod运行，Pod处于Running状态并不能代表应用部署成功。Deployment对Pod副本数量以及应用容器的Health Probe做了设置，以确保进程启动以及应用运行成功。只有当新建ReplicaSet所属的Pod运行成功，并且副本数量达到Deployment设置ReplicaSet的数量，旧的ReplicaSet管理的Pod不再承接任何负载或被销毁时，Deployment达到预期部署状态，才能代表部署成功。"}]}]}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"构造器InitContainer"}]},{"type":"bulletedlist","content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"初始化在OOP中，比如Java类定义，是构造器。封装了对象使用之前必要的初始化操作。初始化容器InitContainer，InitContainer是Pod级别的初始化，同理，是Pod->spec的一类特殊的容器模版定义，隔离了主应用容器进程与初始化操作，确保初始化操作能够在应用容器启动之前完成。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在容器级别也可以完成初始化，利用容器镜像模版Dockerfile->ENTERPOINT定义。容器级别的初始化影响范围是该容器镜像定义内部，而Pod级别初始化操作InitContainer是对Pod内的全部容器组定义。一般情况下，容器初始化更多是Devops关心，与应用开发者工作相关性不大。InitContainer可以将容器模版定义从开发运行周期上完全隔离。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Pod级别初始化操作InitContainer的好处是可以统一设置访问共享Volume的访问权限；在应用服务启动之前准备好依赖的组件或者数据；验证应用服务运行的依赖运行健康等。在应用服务主容器进程启动之前，确保前置条件准备完备，进而确保应用服务能够运行成功。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"处于访问控制等安全性考虑，一般不建议在应用容器镜像定义里开放Pod共享资源的访问权限。我们尽量将Pod的共享资源管理操作留给编排平台设置与控制，与应用服务本身隔离。最大化的确保应用服务本身不带有平台依赖属性。所以，InitContainer也提升了微服务应用开发的安全性。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"一个Pod模版可定义多个InitContainer以及多个应用容器。K8S确保InitContainer按定义顺序依次执行初始化操作，在应用容器启动前执行完毕，而应用容器启动是并行的。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"InitContainer与一般的应用容器基本相同，但是，InitContainer一般为Completed，不会存在Failed终结状态，因为当InitContainer执行失败会直接导致Pod重启，重新开始执行InitContainer。"}]}]}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"组合模式Sidecar"}]},{"type":"bulletedlist","content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在前文我们把容器镜像和容器类比OOP的类和对象，因为容器镜像定义了一个职责单一的应为微服务。那如果在应服务运行时，我们需要扩展或添加一些旁路操作，此时，类似OOP的组合设计模式，我们其实可以直接整合另一个容器镜像定义到同一个Pod，这就是Sidecar。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Sidecar保证了应用容器的职责单一，同时，也能在Pod级别为其添加更新数据，配置文件，静态资源或者采集日志数据这种能够独立复用的旁路操作集合。Sidecar能够通过组合多个职责单一的容器，提供一个功能完备，具备上线能力的应用微服务，同时，确保开发团队只用考虑业务应用功能本身。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Sidecar与InitContainer是两种不同的容器定义。首先InitContainer定义的是Pod级别的初始化操作集合，必须在所有应用容器启动之前执行完毕，具有严格的执行顺序。Sidecar与应用容器执行顺序没有严格控制，两者通常是同时运行在一个Pod内，共享Pod资源，共同完成Pod暴露的服务能力。"}]}]}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"配置管理ConfigMap/Secret"}]},{"type":"bulletedlist","content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"应用开发的12原则中有一条是在环境中存储配置。配置信息与应用隔离，可以通过环境变量来存储应用的配置信息。环境变量具有全局性，可将其在应用进程运行时加载。当配置信息较大时，利用环境变量传递配置信息就不是什么好办法了。Java-Springboot应用提供了Profile文件，记录和保存应用相关的配置信息，开发者可以依据环境区分不同的Profile文件。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在K8S里配置管理ConfigMap和Secret同时支持环境变量Key/Value形式和应用配置Profile文件形式。环境变量的Key一般是全大写字母字母表示；应用配置Profile文件是小写字母表示额文件名。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"K8S提供Secret资源对象来配置敏感数据。比如，数据库链接的用户名，密码等。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"ConfigMap与Secret对象通过Volume挂载到Pod里，所以该配置信息被Pod内的容器组共享。ConfigMap与Secret的数据存储上限为1MB，故当应用配置文件过大，可考虑使用InitContainer初始化一个配置管理容器在同一个Pod下。在应用容器启动前，传递应用配置到应用指定挂载目录，从而更新应用配置信息。"}]}]}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"异步/并发执行Job/Cronjob"}]},{"type":"bulletedlist","content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"应用开发过程中，常会面对批处理任务/定时任务需求。目前流行的应用框架，比如Java的Spring-Batch或者Python的Celery都可以实现异步任务/定时任务。但是这种应用级别的实现方法，在云原生中，会使应用服务实现的很重，比如异步任务通常要求所属应用满足高可用，资源弹性伸缩以及故障自愈。这些特性都是K8S平台天然自带的，可以考虑将应用的异步任务实现委托给K8S的Job/Cronjob控制器对象。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"K8S的Job控制器对象，类似Deployment，是创建与管理Pod生命周期的一种实现方式。与Deployment不同的是，Job控制的Pod是运行结束就终止，即Pod的终态为Completed。Job的Pod默认不会直接销毁，主要目的是提供查看任务运行日志结果。"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"K8S的Cronjob控制器对象，顾名思义，在Job对象之上组合了定时触发事件逻辑。主要使用的场景包括但不限于文件传输，发送邮件或者短信通知，以及备份与定时清理过期备份等。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"结语"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"笔者整理了一部分K8S基础知识点的初衷是为了审视一下K8S这个庞大的技术栈里开发者掌握和使用K8S所要了解的最小知识点集合。笔者相信未来的应用都是建立在云之上，所以不论是哪个角色，都得掌握必要的K8S知识点才能流畅地开启云原生开发之旅。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"以上很多内容都是笔者在学习"},{"type":"link","attrs":{"href":"https://time.geekbang.org/column/intro/116","title":""},"content":[{"type":"text","text":"https://time.geekbang.org/column/intro/116"}]},{"type":"text","text":"以及Kubernetes-patterns"},{"type":"link","attrs":{"href":"https://developers.redhat.com/blog/2020/05/11/top-10-must-know-kubernetes-design-patterns/","title":""},"content":[{"type":"text","text":"https://developers.redhat.com/blog/2020/05/11/top-10-must-know-kubernetes-design-patterns/"}]},{"type":"text","text":"时的学习心得和读书体会，收益于大师们对K8S技术栈多种角度的解读与梳理。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"最后这次比较系统的学习梳理的契机也源于团队对于K8S技术的重视，特别感谢团队领导的重视与支持。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":1}},{"type":"heading","attrs":{"align":null,"level":2}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":1}}]}