Flink1.10任务提交流程分析(二)

在Flink1.10任务提交流程分析(一)中分析了从flink run开始到任务提交到集群前的流程分析，对于不同的提交模式Flink中使用不同的PipelineExecutor，本篇基于yarn-per-job模式分析向yarn-cluster提交任务的流程。(注：基于1.10.1分析)

YarnJobClusterExecutor

接着上篇的分析，任务最终提交是交给PipelineExecutor来execute，PipelineExecutor的选择是根据不同的提交模式来决定即execution.target参数来决定，对于yarn-per-job会选择YarnJobClusterExecutor类型的executor。

public class YarnJobClusterExecutor extends AbstractJobClusterExecutor<ApplicationId, YarnClusterClientFactory> {
   public static final String NAME = "yarn-per-job";
   public YarnJobClusterExecutor() {
      super(new YarnClusterClientFactory());
   }
}

其实现比较简单，比较重要其构造器中YarnClusterClientFactory，用于创建YarnClusterDescriptor，包含了yarn客户端YarnClient、yarn配置、提交yarn的队列等一些提交yarn的信息。它继承了AbstractJobClusterExecutor 抽象任务提交executor，execute也是由AbstractJobClusterExecutor来执行：

public class AbstractJobClusterExecutor<ClusterID, ClientFactory extends ClusterClientFactory<ClusterID>> implements PipelineExecutor {


   private static final Logger LOG = LoggerFactory.getLogger(AbstractJobClusterExecutor.class);
   //代表的就是YarnClusterClientFactory
   private final ClientFactory clusterClientFactory;


   public AbstractJobClusterExecutor(@Nonnull final ClientFactory clusterClientFactory) {
      this.clusterClientFactory = checkNotNull(clusterClientFactory);
   }


   //执行任务提交
   //pipeline 代表StreamGraph
   public CompletableFuture<JobClient> execute(@Nonnull final Pipeline pipeline, @Nonnull final Configuration configuration) throws Exception {
      //将StreamGraph转换为JobGraph
      final JobGraph jobGraph = ExecutorUtils.getJobGraph(pipeline, configuration);
     //创建提交任务的一些信息：YarnClusterDescriptor
      try (final ClusterDescriptor<ClusterID> clusterDescriptor = clusterClientFactory.createClusterDescriptor(configuration)) {
        //将配置信息封装在ExecutionConfigAccessor中
         final ExecutionConfigAccessor configAccessor = ExecutionConfigAccessor.fromConfiguration(