【深入浅出 Yarn 架构与实现】3-2 Yarn Client 编写 _
作者：互联网
1、创建并启动 Client
YarnClient 内容通过 ApplicationClientProtocol 与 ResourceManager 通信，向 RM 的ApplicationsManager 申请 Application。
跟踪进去可以在 YarnClientImpl 找到 rpc：this.rmClient = (ApplicationClientProtocol)ClientRMProxy.createRMProxy(this.getConfig(), ApplicationClientProtocol.class);
YarnClient yarnClient = YarnClient.createYarnClient();
yarnClient.init(conf);
yarnClient.start();
2、通过YarnClient 创建 Application
GetNewApplicationResponse 中除了包含 ApplicationId，还包括集群最大/最小资源，给任务启动设置的资源作参考。
YarnClientApplication app = yarnClient.createApplication();
GetNewApplicationResponse appResponse = app.getNewApplicationResponse();
3、关键：完善 ApplicationSubmissionContext
需要在 ApplicationSubmissionContext 中定义 RM 启动 AM 时所需的全部信息，主要包括：
- app 信息：id，name
- 队列、优先级信息
- 提交用户
- ContainerLaunchContext：定义 AM 启动所需信息
- RECT
- Resources (binaries, jars, files etc.)：其中包括 Application master jar
- Environment settings (CLASSPATH etc.)
- Command to be executed
- security Tokens
- RECT
// 3 完善 ApplicationSubmissionContext 所需内容
ApplicationSubmissionContext appContext = app.getApplicationSubmissionContext();
ApplicationId applicationId = appContext.getApplicationId();
// 3.1 设置application name
appContext.setApplicationName("my-test-app");
// 3.2 设置ContainerLaunchContext
// localResources, env, commands 等
// application master 的 jar 放到 localResources 中
// 这部分较长省略，请到代码中查看
ContainerLaunchContext amContainerCtx = createAMContainerLaunchContext(
conf, app.getApplicationSubmissionContext().getApplicationId());
appContext.setAMContainerSpec(amContainerCtx);
// 3.3 设置优先级
Priority pri = Priority.newInstance(0);
appContext.setPriority(pri);
// 3.4 设置队列
appContext.setQueue("default");
// 3.5 设置 am 资源
int amMemory = 2048;
int amVCores = 2;
Resource capability = Resource.newInstance(amMemory, amVCores);
appContext.setResource(capability);
4、提交 Application
提交后，RM 接收到 Application，根据资源请求分配容器，最终将 AM 启动在容器中。
这里交给 YarnClientImpl 执行 rmClient.submitApplication(request)，通过 RPC ApplicationClientProtocol 提交到 RM
ApplicationId appId = yarnClient.submitApplication(appContext);
5、获取任务进度信息
ApplicationReport report = yarnClient.getApplicationReport(appId);
log.info("Got application report " +
", clientToAMToken=" + report.getClientToAMToken()
+ ", appDiagnostics=" + report.getDiagnostics()
+ ", appMasterHost=" + report.getHost()
+ ", appQueue=" + report.getQueue()
+ ", appMasterRpcPort=" + report.getRpcPort()
+ ", appStartTime=" + report.getStartTime()
+ ", yarnAppState=" + report.getYarnApplicationState().toString()
+ ", distributedFinalState=" + report.getFinalApplicationStatus().toString()
+ ", appTrackingUrl=" + report.getTrackingUrl()
+ ", appUser=" + report.getUser());
6、kill Application
当 Application 运行了过长的时间或者其他的原因，client 可以 kill application。
流程是：client 像 RM 发送 kill 信号，再传递给 AM
yarnClient.killApplication(appId);