云原生之旅 - 11)基于 Kubernetes 动态伸缩 Jenkins Build Agents
阅读原文时间:2023年07月08日阅读:1

上一篇文章 云原生之旅 - 10)手把手教你安装 Jenkins on Kubernetes 我们介绍了在 Kubernetes 上安装 Jenkins,本文介绍下如何设置k8s pod作为Jenkins 构建job的 agent。

Jenkins master 和 agent 均以 pod 的形式运行在 Kubernetes 节点上。Master 运行在其中一个节点上,其配置数据 Jenkins home 使用存储卷挂载,master pod重启不会导致数据丢失。agent 运行在各个节点上,根据需求动态创建并自动释放。这样做的好处很多,比如高可用,高伸缩性,资源利用率高。

关键词:Jenkins on Kubernetes 实践,Jenkins 和 Kubernetes,在Kubernetes上安装Jenkins,Jenkins 高可用安装,Jenkins 动态伸缩构建, Kubernetes Pod as Jenkins build agent

  1. 已搭建 Jenkins master on kubernetes 云原生之旅 - 10)手把手教你安装 Jenkins on Kubernetes
  2. 准备一个 Service Account,对目标 cluster 具有k8s admin权限,以便部署。
  3. 防火墙已开通 Jenkins 出站到Docker hub,方便 push/pull image
  4. 防火墙已开通 Jenkins 到 目标 cluster,以便部署。
  • Kubernetes Plugin
  • Google Kubernetes Engine Plugin (我的例子是部署到 GKE cluster)

Manage Nodes and Clouds

1. Go to `Manage Jenkins` –> `Manage Nodes and Clouds`

2. Click `Configure Clouds`

3. Add a new Cloud select `Kubernetes`

4. Click `Kubernetes Cloud Detail

5. Enter `jenkins` namespace in `Kubernetes Namespace` field

6. Click `Test Connection` --> result show `Connected to Kubernetes v1.22.12-gke.2300`

7. Click `Save`

8. Enter `http://jenkins-service.jenkins.svc.cluster.local:8080` in `Jenkins URL` field

9. Enter `jenkins-agent:50000` in `Jenkins tunnel` field

10. Click `Add Pod Template` then `Pod Template Details`

11. Input `Name`=`jenkins-agent`, `Namespace`=`jenkins`, `Labels`=`kubeagent`

12. (Optional) 如果不添加 container template, the Jenkins Kubernetes plugin will use the default JNLP image from the Docker hub to spin up the agents.

如果你要覆盖默认的jnlp image 可以 Click `Add Container` to add Container Template,

输入 `Name`=`jnlp`, `Docker Image`=`your_registry/jenkins/inbound-agent:4.11-1-jdk11`

Ensure that you remove the sleep and 9999999 default argument from the container template.

Manage Credentials

  • Add `Usernames with password` for docker hub account/pwd,比如 wade_test_dockerhub
  • Add `Google Service Account from private key` 比如 gcp_sa_json_key

Credentials 会在Jenkinsfile里面用到。

### 本文首发于博客园 https://www.cnblogs.com/wade-xu/p/16863955.html

Go to Jenkins home –> New Item and create a freestyle project,命名为 quick-test

在 job description 部分, add the label `kubeagent` for `Restrict where this project can be run`.

这个label 和我们上面创建 pod template时用的label一致. 这样的话 Jenkins就知道用哪个 pod template 作为 agent container.

随便添加一个shell 作为build steps

点Build Now

 查看Console Output

Agent jenkins-agent-l7hw9 is provisioned from template jenkins-agent

……

Building remotely on jenkins-agent-l7hw9 (kubeagent) in workspace /home/jenkins/agent/workspace/quick-test
[quick-test] $ /bin/sh -xe /tmp/jenkins17573873264046707236.sh

  • echo test pipeline
    test pipeline
    Finished: SUCCESS

### 本文首发于博客园 https://www.cnblogs.com/wade-xu/p/16863955.html

CI

接着我们用 Jenkinsfile 写一个 Declarative pipeline - build/push docker image 到docker hub

首先需要定义一个 pod.yaml 作为启动 agent 的container

kind: Pod
spec:
containers: # list of containers that you want present for your build, you can define a default container in the Jenkinsfile
- name: maven
image: maven:3.5.4-jdk-8-slim
command: ["tail", "-f", "/dev/null"] # this or any command that is bascially a noop is required, this is so that you don't overwrite the entrypoint of the base container
imagePullPolicy: Always # use cache or pull image for agent
resources: # request and limit the resources your build contaienr
requests:
memory: 4Gi
cpu: 2
limits:
memory: 4Gi
cpu: 2
volumeMounts:
- mountPath: /root/.m2 # maven .m2 cache directory
name: maven-home
- name: git
image: bitnami/git:2.38.1
imagePullPolicy: IfNotPresent
command: ["tail", "-f", "/dev/null"]
resources: # limit the resources your build contaienr
limits:
cpu: 100m
memory: 256Mi
- name: kubectl-kustomize
image: line/kubectl-kustomize:1.25.3-4.5.7
imagePullPolicy: IfNotPresent
command: ["tail", "-f", "/dev/null"]
resources: # limit the resources your build contaienr
limits:
cpu: 100m
memory: 256Mi
- name: docker
image: docker:18.06.1
command: ["tail", "-f", "/dev/null"]
imagePullPolicy: Always
volumeMounts:
- name: docker
mountPath: /var/run/docker.sock # We use the k8s host docker engine
volumes:
- name: docker
hostPath:
path: /var/run/docker.sock
- name: maven-home
persistentVolumeClaim:
claimName: maven-repo-storage

build-pod.yaml

在Jenkinsfile里面定义agent 使用这个yaml file

agent {
kubernetes {
idleMinutes 3 // how long the pod will live after no jobs have run on it
yamlFile './build-pod.yaml' // path to the pod definition relative to the root of our project
defaultContainer 'docker' // define a default container if more than a few stages use it, otherwise default to jnlp container
}

下面步骤是 docker login/build/tag/push

  environment {

    DOCKER_HUB_REGISTRY='https://index.docker.io/v1/'

    DOCKER_HUB_CREDS = credentials('wade_test_dockerhub')

  }

stage('Build and Push Docker Image') {
steps {
script {
dir(dir_path) {
container('docker') {
// docker login, Using single-quotes instead of double-quotes when referencing these sensitive environment variables prevents this type of leaking.
sh 'echo $DOCKER_HUB_CREDS_PSW | docker login -u $DOCKER_HUB_CREDS_USR --password-stdin $DOCKER_HUB_REGISTRY'
// build image with git tag
sh """
docker build -t $PROJECT_IMAGE_WITH_TAG .
docker tag $PROJECT_IMAGE_WITH_TAG $DOCKER_HUB_CREDS_USR/$PROJECT_IMAGE_WITH_TAG
"""

            // push image\_tag to docker hub  
            sh """  
            docker push $DOCKER\_HUB\_CREDS\_USR/$PROJECT\_IMAGE\_WITH\_TAG  
            """  
        }  
      }  
    }  
  }  
}

我这里没有选择用 docker.withRegistry

docker.withRegistry("$DOCKER_HUB_REGISTRY", "$DOCKER_HUB_CREDENTIAL") {}

因为会有不安全的log提示

WARNING! Using --password via the CLI is insecure. Use --password-stdin.

CI + Kustomize + CD

这个例子是上面的 CI 之后 加上 - 利用 Kustomize build K8S resource manifests 然后 CD 到一个 Cluster

Kustomize 可以参考 云原生之旅 - 6)不能错过的一款 Kubernetes 应用编排管理神器 Kustomize

// assume your k8s manifests in another repo, mine is same repo, just in order to show git clone step  
stage('Checkout K8S manifests') {  
  steps {  
    script {  
      dir(dir\_path) {  
        container('git') {  
          if (! fileExists('learning\_by\_doing/README.md')) {  
            sh """  
            git clone https://github.com/wadexu007/learning\_by\_doing.git  
            ls -lhrt  
            """  
          } else {  
              sh 'echo manifes repo already exist.'  
          }  
        }  
      }  
    }  
  }  
}

stage('Build manifests with Kustomize') {  
  steps {  
    script {  
      dir(dir\_path) {  
        container('kubectl-kustomize') {  
            sh """  
            cd learning\_by\_doing/Kustomize/demo-manifests/services/demo-app/dev/  
            kustomize edit set image $DOCKER\_HUB\_CREDS\_USR/$PROJECT\_IMAGE\_WITH\_TAG  
            kustomize build > $WORKSPACE/$dir\_path/deployment.yaml  
            """  
        }  
      }  
    }  
  }  
}

stage('Deploy to GKE test cluster') {  
        environment{  
            PROJECT\_ID = 'xperiences-eng-cn-dev'  
    CLUSTER\_NAME = 'xpe-spark-test-gke'  
    REGION = 'asia-east2'  
    CREDENTIALS\_ID = 'gcp\_sa\_json\_key'  
  }  
  steps {  
    script {  
      dir(dir\_path) {  
        container('kubectl-kustomize') {  
            sh """  
            chown 1000:1000 deployment.yaml  
            echo start to deploy to cluster $CLUSTER\_NAME  
            """  
            step(\[  
              $class: 'KubernetesEngineBuilder',  
              projectId: env.PROJECT\_ID,  
              clusterName: env.CLUSTER\_NAME,  
              location: env.REGION,  
              manifestPattern: 'deployment.yaml',  
              credentialsId: env.CREDENTIALS\_ID,  
              verifyDeployments: false\])  
              // verifyDeployments does not work for non-default namespace  
        }  
      }  
    }  
  }  
}

Pipeline: Input Step

这个例子是利用 Jenkins pipeline的 Input step 来做一个人工介入Approve的步骤。 然后再来一个多cluster 部署,选不同region 部署到不同的cluster的示例。

stage('Wait for SRE Approval') {  
  steps {  
    timeout(time:72, unit:'HOURS') {  
      input message: "Approved Prod deployment?", submitter: 'sre-team'  
    }  
  }  
}

// deployment to multipe k8s clusters  
stage('Deploy to GKE Prod cluster') {  
        environment{  
            PROJECT\_ID = 'sre-cn-dev'  
    CREDENTIALS\_ID = 'gcp\_sa\_json\_key'  
    CLUSTER\_COMMON\_NAME = 'demo-gke-prod'  
  }  
  steps {  
    script {  
      env.REGION = input message: 'Choose which region you want to deploy?',  
                         parameters: \[choice(name: 'Region',  
                                            description: 'Select Region to Deloy',  
                                            choices: \['europe-west1', 'us-central1'\])  
                                      \]  
      dir(dir\_path) {  
        if ( env.REGION == "europe-west1" ) {  
          def eu\_cluster\_name = env.CLUSTER\_COMMON\_NAME + "-eu"  
          container('kubectl-kustomize') {  
              sh "echo deploy to cluster $eu\_cluster\_name in region: $REGION"  
          }  
        }  
        if ( env.REGION == "us-central1" ) {  
          def us\_cluster\_name = env.CLUSTER\_COMMON\_NAME + "-us"  
          container('kubectl-kustomize') {  
              sh "echo deploy to cluster $us\_cluster\_name in region: $REGION"  
          }  
        }  
      }  
    }  
  }  
}

所有例子均在我的 github repo

### 本文首发于博客园 https://www.cnblogs.com/wade-xu/p/16863955.html

现在你可以创建一个 Pipeline 或者 Multibranch Pipeline job 来测试。

Repository URL = `https://github.com/wadexu007/learning_by_doing`

Script Path, e.g. `Jenkins/k8s_pod_as_build_agent/demo-app-java/Jenkinsfile`

你会看到每启动一个job 都会相应的产生一个pod 来作为Jenkins agent运行,结束后根据idleMinutes自动释放。

 

如果你已经成功创建并测试 CI/CD pipeline,可以继续加强,比如加上 Post notifications

  • 设置 resource requests and limits on each container in your Pod
  • 如果使用maven 构建 java项目,.m2 cache目录需要 mount 出来,这样加快后面的maven build速度。
  • 使用 Jenkins Shared Libraries 抽取Pipeline的共用代码
  • 在容器里构建容器化应用(Run docker in docker) 我的例子是通过 mount docker.sock 利用k8s 主机 docker engine来实现的,这种方式需要 privileges mode 不安全,推荐使用Kaniko,下一篇文章会介绍。

感谢阅读,如果您觉得本文的内容对您的学习有所帮助,您可以打赏和推荐,您的鼓励是我创作的动力。