DevOps基于k8s发布系统的实现

综述

首先,本篇文章所介绍的内容,已经有完整的实现,可以参考这里。
在微服务、DevOps和云平台流行的当下,使用一个高效的持续集成工具也是一个非常重要的事情。虽然市面上目前已经存在了比较成熟的自动化构建工具,比如jekines,还有一些商业公司推出的自动化构建工具,但他们都不能够很好的和云环境相结合。那么究竟该如何实现一个简单、快速的基于云环境的自动化构建系统呢?我们首先以一个Springboot应用为例来介绍一下整体的发布流程,然后再来看看具体如何实现。发布的步骤大体如下:
1.首先从代码仓库下载代码,比如Gitlab、GitHub等;
2.接着是进行打包,比如使用Maven、Gradle等;
3.如果要使用k8s作为编排,还需要把步骤2产生的包制作成镜像,比如用Docker等;
4.上传步骤3的镜像到远程仓库,比如Harhor、DockerHub等;
5.最后,下载镜像并编写Deployment文件部署到k8s集群;
如图1所示:
DevOps基于k8s发布系统的实现_第1张图片
图1

从以上步骤可以看出,发布过程中需要的工具和环境至少包括:代码仓库(Gitlab、GitHub等)、打包环境(Maven、Gradle等)、镜像制作(Docker等)、镜像仓库(Harbor、DockerHub等)、k8s集群等;此外,还包括发布系统自身的数据存储等。
可以看出,整个流程里依赖的环境很多,如果发布系统不能与这些环境解耦,那么要想实现一个安装简单、功能快速的系统没有那么容易。那么有没有合理的解决方案来实现与这些环境的解耦呢?答案是有的,下面就分别介绍。

代码仓库

操作代码仓库,一般系统提供的都有对应Restful API,以GitLab系统提供的Java客户端为例,如下代码:

<dependency>
	<groupId>org.gitlab4jgroupId>
	<artifactId>gitlab4j-apiartifactId>
	<version>4.17.0version>
dependency>

比如,我们想获取某个项目的分支列表,如下代码所示:

public List<Branch> branchList(CodeRepo codeRepo, BranchListParam param) {
	GitLabApi gitLabApi = gitLabApi(codeRepo);
	List<Branch> list = null;
	try {
		list = gitLabApi.getRepositoryApi().getBranches(param.getProjectIdOrPath(), param.getBranchName());
	} catch (GitLabApiException e) {
		LogUtils.throwException(logger, e, MessageCodeEnum.PROJECT_BRANCH_PAGE_FAILURE);
	} finally {
		gitLabApi.close();
	}
}

private GitLabApi gitLabApi(CodeRepo codeRepo) {
	GitLabApi gitLabApi = new GitLabApi(codeRepo.getUrl(), codeRepo.getAuthToken());
	gitLabApi.setRequestTimeout(1000, 5 * 1000);
	try {
		gitLabApi.getVersion();
	}catch(GitLabApiException e) {
		//如果token无效,则用账号登录
		if(e.getHttpStatus() == 401 && !StringUtils.isBlank(codeRepo.getAuthUser())) {
			gitLabApi = new GitLabApi(codeRepo.getUrl(), codeRepo.getAuthUser(), codeRepo.getAuthPassword());
			gitLabApi.setRequestTimeout(1000, 5 * 1000);
		}
	}
	
	return gitLabApi;
}

打包环境

我们以Maven为例进行说明,一般情况下,我们使用Maven打包时,需要首先安装Maven环境,接着引入打包插件,然后使用mvn clean package命令就可以打包了。比如springboot自带插件:

<plugin>
	<groupId>org.springframework.bootgroupId>
	<artifactId>spring-boot-maven-pluginartifactId>
	<version>2.5.6version>
	<configuration>
		<classifier>executeclassifier>
		<mainClass>com.test.ApplicationmainClass>
	configuration>
	<executions>
		<execution>
			<goals>
				<goal>repackagegoal>
			goals>
		execution>
	executions>
plugin>

再比如,通用的打包插件:

<plugin>
	<groupId>org.apache.maven.pluginsgroupId>
	<artifactId>maven-assembly-pluginartifactId>
	<version>3.8.2version>
	<configuration>
		<appendAssemblyId>falseappendAssemblyId>
		<descriptors>
			<descriptor>src/main/resources/assemble.xmldescriptor>
		descriptors>
		<outputDirectory>../targetoutputDirectory>
	configuration>
	<executions>
		<execution>
			<id>make-assemblyid>
			<phase>packagephase>
			<goals>
				<goal>singlegoal>
			goals>
		execution>
	executions>
plugin>

等等。然后再通过运行mvn clean package命令进行打包。那么,在打包时如果要去除对maven环境的依赖,该如何实现呢?
可以使用嵌入式maven插件maven-embedder来实现。
具体可以这样来做,首先在平台项目里引入依赖,如下:

<dependency>
	<groupId>org.apache.mavengroupId>
	<artifactId>maven-embedderartifactId>
	<version>3.8.1version>
dependency>
<dependency>
	<groupId>org.apache.mavengroupId>
	<artifactId>maven-compatartifactId>
	<version>3.8.1version>
dependency>
<dependency>
	<groupId>org.apache.maven.resolvergroupId>
	<artifactId>maven-resolver-connector-basicartifactId>
	<version>1.7.1version>
dependency>
<dependency>
	<groupId>org.apache.maven.resolvergroupId>
	<artifactId>maven-resolver-transport-httpartifactId>
	<version>1.7.1version>
dependency>

运行如下代码,就可以对项目进行打包了:

String[] commands = new String[] { "clean", "package", "-Dmaven.test.skip" };
String pomPath = "D:/hello/pom.xml";
MavenCli cli = new MavenCli();
try {
	cli.doMain(commands, pomPath, System.out, System.out);
} catch (Exception e) {
	e.printStackTrace();
}

但是,一般情况下,我们通过maven的settings文件还会做一些配置,比如配置工作目录、nexus私服地址、Jdk版本、编码方式等等,如下:


<settings xmlns="http://maven.apache.org/SETTINGS/1.0.0"
	xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
	xsi:schemaLocation="http://maven.apache.org/SETTINGS/1.0.0 http://maven.apache.org/xsd/settings-1.0.0.xsd">
	<localRepository>C:/m2/repositorylocalRepository>
	<profiles>
		<profile>
			<id>myNexusid>
			<repositories>
				<repository>
					<id>nexusid>
					<name>nexusname>
					<url>https://repo.maven.apache.org/maven2url>
					<releases>
						<enabled>trueenabled>
					releases>
					<snapshots>
						<enabled>trueenabled>
					snapshots>
				repository>
			repositories>
			<pluginRepositories>
				<pluginRepository>
					<id>nexusid>
					<name>nexusname>
					<url>https://repo.maven.apache.org/maven2url>
					<releases>
						<enabled>trueenabled>
					releases>
					<snapshots>
						<enabled>trueenabled>
					snapshots>
				pluginRepository>
			pluginRepositories>
		profile>

		<profile>
			<id>java11id>
			<activation>
				<activeByDefault>trueactiveByDefault>
				<jdk>11jdk>
			activation>
			<properties>
				<maven.compiler.source>11maven.compiler.source>
				<maven.compiler.target>11maven.compiler.target>
				<maven.compiler.compilerVersion>11maven.compiler.compilerVersion>
				<project.build.sourceEncoding>UTF-8project.build.sourceEncoding>
				<project.build.outputEncoding>UTF-8project.build.outputEncoding>
			properties>
		profile>
	profiles>
	<activeProfiles>
		<activeProfile>myNexusactiveProfile>
	activeProfiles>
settings>

通过查看MavenCli类发现,doMain(CliRequest cliRequest)方法有比较丰富的参数,CliRequest的代码如下:

package org.apache.maven.cli;

public class CliRequest
{
    String[] args;

    CommandLine commandLine;

    ClassWorld classWorld;

    String workingDirectory;

    File multiModuleProjectDirectory;

    boolean debug;

    boolean quiet;

    boolean showErrors = true;

    Properties userProperties = new Properties();

    Properties systemProperties = new Properties();

    MavenExecutionRequest request;

    CliRequest( String[] args, ClassWorld classWorld )
    {
        this.args = args;
        this.classWorld = classWorld;
        this.request = new DefaultMavenExecutionRequest();
    }

    public String[] getArgs()
    {
        return args;
    }

    public CommandLine getCommandLine()
    {
        return commandLine;
    }

    public ClassWorld getClassWorld()
    {
        return classWorld;
    }

    public String getWorkingDirectory()
    {
        return workingDirectory;
    }

    public File getMultiModuleProjectDirectory()
    {
        return multiModuleProjectDirectory;
    }

    public boolean isDebug()
    {
        return debug;
    }

    public boolean isQuiet()
    {
        return quiet;
    }

    public boolean isShowErrors()
    {
        return showErrors;
    }

    public Properties getUserProperties()
    {
        return userProperties;
    }

    public Properties getSystemProperties()
    {
        return systemProperties;
    }

    public MavenExecutionRequest getRequest()
    {
        return request;
    }

    public void setUserProperties( Properties properties ) 
    {
        this.userProperties.putAll( properties );      
    }
}

可以看出,这些参数非常丰富,也许可以满足我们的需求,但是CliRequest只有一个默认修饰符的构造方法,也就说只有位于org.apache.maven.cli包下的类才有访问CliRequest构造方法的权限,我们可以在平台项目里新建一个包org.apache.maven.cli,然后再创建一个类(如:DefaultCliRequest)继承自CliRequest,然后实现一个public的构造方法,就可以在任何包里使用该类了,如下代码:

package org.apache.maven.cli;

import org.codehaus.plexus.classworlds.ClassWorld;

public class DefaultCliRequest extends CliRequest{

	public DefaultCliRequest(String[] args, ClassWorld classWorld) {
		super(args, classWorld);
	}
	
	public void setWorkingDirectory(String directory) {
		this.workingDirectory = directory;
	}
}

定义好参数类型DefaultCliRequest后,我们再来看看打包的代码:

public void doPackage() {
	String[] commands = new String[] { "clean", "package", "-Dmaven.test.skip" };
	DefaultCliRequest request = new DefaultCliRequest(commands, null);
	request.setWorkingDirectory("D:/hello/pom.xml");

	Repository repository = new Repository();
	repository.setId("nexus");
	repository.setName("nexus");
	repository.setUrl("https://repo.maven.apache.org/maven2");
	RepositoryPolicy policy = new RepositoryPolicy();
	policy.setEnabled(true);
	policy.setUpdatePolicy("always");
	policy.setChecksumPolicy("fail");
	repository.setReleases(policy);
	repository.setSnapshots(policy);

	String javaVesion = "11";
	Profile profile = new Profile();
	profile.setId("java11");
	Activation activation = new Activation();
	activation.setActiveByDefault(true);
	activation.setJdk(javaVesion);
	profile.setActivation(activation);
	profile.setRepositories(Arrays.asList(repository));
	profile.setPluginRepositories(Arrays.asList(repository));

	Properties properties = new Properties();
	properties.put("java.home", "D:/java/jdk-11.0.16.2");
	properties.put("java.version", javaVesion);
	properties.put("maven.compiler.source", javaVesion);
	properties.put("maven.compiler.target", javaVesion);
	properties.put("maven.compiler.compilerVersion", javaVesion);
	properties.put("project.build.sourceEncoding", "UTF-8");
	properties.put("project.reporting.outputEncoding", "UTF-8");
	profile.setProperties(properties);
	MavenExecutionRequest executionRequest = request.getRequest();
	executionRequest.setProfiles(Arrays.asList(profile));

	MavenCli cli = new MavenCli();
	try {
		cli.doMain(request);
	} catch (Exception e) {
		e.printStackTrace();
	}
}

如果需要设置其他参数,也可以通过以上参数自行添加。

镜像制作

一般情况下,我们在Docker环境中通过Docker命令来制作镜像,过程如下:
1.首先编写Dockerfile文件;
2.通过docker build制作镜像;
3.通过docker push上传镜像;
可以看出,如果要使用docker制作镜像的话,必须要有docker环境,而且需要编写Dockerfile文件。当然,也可以不用安装docker环境,直接使用doker的远程接口:post/build。但是,在远程服务器中仍然需要安装doker环境和编写Dockerfile。在不依赖Docker环境的情况下,仍然可以制作镜像,下面就介绍一款工具Jib的用法。
Jib是谷歌开源的一套工具,github地址,它是一个无需Docker守护进程——也无需深入掌握Docker最佳实践的情况下,为Java应用程序构建Docker和OCI镜像, 它可以作为Maven和Gradle的插件,也可以作为Java库。

比如,使用jib-maven-plugin插件构建镜像的代码如下:

<plugin>
	<groupId>com.google.cloud.toolsgroupId>
	<artifactId>jib-maven-pluginartifactId>
	<version>3.3.0version>
	<configuration>
		<from>
			<image>openjdk:13-jdk-alpineimage>
		from>
		<to>
			<image>gcr.io/dhorse/clientimage>
			<tags>
				<tag>102tag>
			tags>
			<auth>
				
				<username>usernameusername>
				<password>passwordpassword>
			auth>
		to>
		<container>
			<ports>
				<port>8080port>
			ports>
		container>
	configuration>
	<executions>
		<execution>
			<phase>packagephase>
			<goals>
				<goal>buildgoal>
			goals>
		execution>
	executions>
plugin>

然后使用命令进行构建:

mvn compile jib:build

可以看出,无需docker环境就可以实现镜像的构建。但是,要想通过平台类型的系统去为每个系统构建镜像,显然通过插件的方式,不太合适,因为需要每个被构建系统引入jib-maven-plugin插件才行,也就是需要改造每一个系统,这样就会带来一定的麻烦。那么有没有不需要改造系统的方式直接进行构建镜像呢?答案是通过Jib-core就可以实现。

首先,在使用Jib-core的项目中引入依赖,maven如下:

<dependency>
	<groupId>com.google.cloud.toolsgroupId>
	<artifactId>jib-coreartifactId>
	<version>0.22.0version>
dependency>

然后就可以直接使用Jib-core的API来进行制作镜像,如下代码:

try {
	JibContainerBuilder jibContainerBuilder = null;
	if (StringUtils.isBlank(context.getProject().getBaseImage())) {
		jibContainerBuilder = Jib.fromScratch();
	} else {
		jibContainerBuilder = Jib.from(context.getProject().getBaseImage());
	}
	//连接镜像仓库5秒超时
	System.setProperty("jib.httpTimeout", "5000");
	System.setProperty("sendCredentialsOverHttp", "true");
	String fileNameWithExtension = targetFiles.get(0).toFile().getName();
	List<String> entrypoint = Arrays.asList("java", "-jar", fileNameWithExtension);
	RegistryImage registryImage = RegistryImage.named(context.getFullNameOfImage()).addCredential(
			context.getGlobalConfigAgg().getImageRepo().getAuthUser(),
			context.getGlobalConfigAgg().getImageRepo().getAuthPassword());
	jibContainerBuilder.addLayer(targetFiles, "/")
		.setEntrypoint(entrypoint)
		.addVolume(AbsoluteUnixPath.fromPath(Paths.get("/etc/localtime")))
		.containerize(Containerizer.to(registryImage)
				.setAllowInsecureRegistries(true)
				.addEventHandler(LogEvent.class, logEvent -> logger.info(logEvent.getMessage())));
} catch (Exception e) {
	logger.error("Failed to build image", e);
	return false;
}

其中,targetFiles是要构建镜像的目标文件,比如springboot打包后的jar文件。
通过Jib-core,可以很轻松的实现镜像构建,而不需要依赖任何其他环境,也不需要被构建系统做任何改造,非常方便。

镜像仓库

类似代码仓库提供的Restful API,也可以通过Restful API来操作镜像仓库,以Harbor创建一个项目为例,代码如下:

public void createProject(ImageRepo imageRepo) {
	String uri = "api/v2.0/projects";
	if(!imageRepo.getUrl().endsWith("/")) {
		uri = "/" + uri;
	}
	HttpPost httpPost = new HttpPost(imageRepo.getUrl() + uri);
	RequestConfig requestConfig = RequestConfig.custom()
			.setConnectionRequestTimeout(5000)
			.setConnectTimeout(5000)
			.setSocketTimeout(5000)
			.build();
	httpPost.setConfig(requestConfig);
	httpPost.setHeader("Content-Type", "application/json;charset=UTF-8");
	httpPost.setHeader("Authorization", "Basic "+ Base64.getUrlEncoder().encodeToString((imageRepo.getAuthUser() + ":" + imageRepo.getAuthPassword()).getBytes()));
	ObjectNode objectNode = JsonUtils.getObjectMapper().createObjectNode();
	objectNode.put("project_name", "dhorse");
	//1:公有类型
	objectNode.put("public", 1);
	httpPost.setEntity(new StringEntity(objectNode.toString(),"UTF-8"));
	try (CloseableHttpResponse response = createHttpClient(imageRepo.getUrl()).execute(httpPost)){
		if (response.getStatusLine().getStatusCode() != 201
				&& response.getStatusLine().getStatusCode() != 409) {
			LogUtils.throwException(logger, response.getStatusLine().getReasonPhrase(),
					MessageCodeEnum.IMAGE_REPO_PROJECT_FAILURE);
		}
	} catch (IOException e) {
		LogUtils.throwException(logger, e, MessageCodeEnum.IMAGE_REPO_PROJECT_FAILURE);
	}
}

k8s集群

同样,k8s也提供了Restful API。同时,官方也提供了各种语言的客户端,下面以Java语言的客户端为例,来创建一个deployment。
首先,引入Maven依赖:

<dependency>
	<groupId>io.kubernetesgroupId>
	<artifactId>client-javaartifactId>
	<version>13.0.0version>
dependency>

然后,使用如下代码:

public boolean createDeployment(DeployContext context) {
	V1Deployment deployment = new V1Deployment();
	deployment.apiVersion("apps/v1");
	deployment.setKind("Deployment");
	deployment.setMetadata(deploymentMetaData(context.getDeploymentAppName()));
	deployment.setSpec(deploymentSpec(context));
	ApiClient apiClient = this.apiClient(context.getCluster().getClusterUrl(),
			context.getCluster().getAuthToken(), 1000, 1000);
	AppsV1Api api = new AppsV1Api(apiClient);
	CoreV1Api coreApi = new CoreV1Api(apiClient);
	String namespace = context.getProjectEnv().getNamespaceName();
	String labelSelector = K8sUtils.getDeploymentLabelSelector(context.getDeploymentAppName());
	try {
		V1DeploymentList oldDeployment = api.listNamespacedDeployment(namespace, null, null, null, null,
				labelSelector, null, null, null, null, null);
		if (CollectionUtils.isEmpty(oldDeployment.getItems())) {
			deployment = api.createNamespacedDeployment(namespace, deployment, null, null, null);
		} else {
			deployment = api.replaceNamespacedDeployment(context.getDeploymentAppName(), namespace, deployment, null, null,
					null);
		}
	} catch (ApiException e) {
		if (!StringUtils.isBlank(e.getMessage())) {
			logger.error("Failed to create k8s deployment, message: {}", e.getMessage());
		} else {
			logger.error("Failed to create k8s deployment, message: {}", e.getResponseBody());
		}
		return false;
	}
	return true;
}

private ApiClient apiClient(String basePath, String accessToken, int connectTimeout, int readTimeout) {
	ApiClient apiClient = new ClientBuilder().setBasePath(basePath).setVerifyingSsl(false)
			.setAuthentication(new AccessTokenAuthentication(accessToken)).build();
	apiClient.setConnectTimeout(connectTimeout);
	apiClient.setReadTimeout(readTimeout);
	return apiClient;
}

至此,关键的技术点已经介绍完了,更多内容,请参考这里

也可以参考其他文章:
《DHorse系列文章之操作手册》
《DHorse系列文章之镜像制作》
《DHorse系列文章之maven打包》
《DHorse系列文章之Dubbo项目解决方案》

你可能感兴趣的:(微服务治理,k8s,运维,kubernetes,DevOps,持续集成,集成工具)