【Flink源码】从StreamExecutionEnvironment.execute看Flink提交过程

env.execute("Order Count");

相信大家对这一行代码都不陌生，其作用是执行 Flink 程序，相当于是一个总开关。
很难想象，那么复杂的 Flink 架构，那么复杂的 Flink 程序仅仅需要这简单的一个函数就能启动，其背后究竟是怎样的过程？

---

execute 与 Flink 执行原理

StreamExecutionEnvironment.java

public JobExecutionResult execute() throws Exception {
    return execute((String) null);
}

execute 方法的参数为 jobName，若未指定则自动赋为 null

/**
* Triggers the program execution. The environment will execute all parts of the program that
* have resulted in a "sink" operation. Sink operations are for example printing results or
* forwarding them to a message queue.
*
* <p>The program execution will be logged and displayed with the provided name
*
* @param jobName Desired name of the job
* @return The result of the job execution, containing elapsed time and accumulators.
* @throws Exception which occurs during job execution.
*/
public JobExecutionResult execute(String jobName) throws Exception {
    final List<Transformation<?>> originalTransformations = new ArrayList<>(transformations);
    StreamGraph streamGraph = getStreamGraph();
    if (jobName != null) {
        streamGraph.setJobName(jobName);
    }

    try {
        return execute(streamGraph);
    } catch (Throwable t) {
        Optional<ClusterDatasetCorruptedException> clusterDatasetCorruptedException =
                ExceptionUtils.findThrowable(t, ClusterDatasetCorruptedException.class);
        if (!clusterDatasetCorruptedException.isPresent()) {
            throw t;
        }

        // Retry without cache if it is caused by corrupted cluster dataset.
        invalidateCacheTransformations(originalTransformations);
        streamGraph = getStreamGraph(originalTransformations);
        return execute(streamGraph);
    }
}

这一大段注释的大意是触发程序执行，环境将执行导致 sink 操作的程序的所有部分。
该方法首先通过 getStreamGraph 方法获取了 StreamGraph 对象。

public StreamGraph getStreamGraph() {
    return getStreamGraph(true);
}

public StreamGraph getStreamGraph(boolean clearTransformations) {
    final StreamGraph streamGraph = getStreamGraph(transformations);
    if (clearTransformations) {
        transformations.clear();
    }
    return streamGraph;
}

由源码可知该方法的主要作用是获取流的执行图，若参数 clearTransformations 为 true（默认为 true）则清空 transformations。
这里的 transformations 是一个 List 的对象，包含一系列流的转换操作，而 Transformation 本身是一个抽象类，用于完成从输入流到输出流的转换，也就是我们常用的 map、filter 等转换算子其底层都是一棵 Transformation 树。任何一个 Flink 程序只要包含流的输入与输出都会存在一棵 Transformation 树。
Flink 程序会基于 Transformation 列表将其转化为 StreamGraph
这里清空 transformations 也就是清除这棵转换树。在完成了到 StreamGraph 的转换后清除树。
我们再继续往下看调用的 getStreamGraph(List<Transformation<?>> transformations)

private StreamGraph getStreamGraph(List<Transformation<?>> transformations) {
    synchronizeClusterDatasetStatus();
    return getStreamGraphGenerator(transformations).generate();
}

private void synchronizeClusterDatasetStatus() {
    if (cachedTransformations.isEmpty()) {
        return;
    }
    Set<AbstractID> completedClusterDatasets =
            listCompletedClusterDatasets().stream()
                    .map(AbstractID::new)
                    .collect(Collectors.toSet());
    cachedTransformations.forEach(
            (id, transformation) -> {
                transformation.setCached(completedClusterDatasets.contains(id));
            });
}

synchronizeClusterDatasetStatus 顾名思义，同步集群数据集状态。
其中，cachedTransformations 是一个 Map<AbstractID, CacheTransformation<?>> 变量，表示集群中各个数据集的缓存转换算子。synchronizeClusterDatasetStatus 方法就是将已完成算子到执行图转化的数据集列表与缓存列表同步。
接下来，调用 getStreamGraphGenerator 生成执行图。

public StreamGraph generateStreamGraph(List<Transformation<?>> transformations) {
    return getStreamGraphGenerator(transformations).generate();
}

private StreamGraphGenerator getStreamGraphGenerator(List<Transformation<?>> transformations) {
    if (transformations.size() <= 0) {
        throw new IllegalStateException(
                "No operators defined in streaming topology. Cannot execute.");
    }

    // We copy the transformation so that newly added transformations cannot intervene with the
    // stream graph generation.
    return new StreamGraphGenerator(
                    new ArrayList<>(transformations), config, checkpointCfg, configuration)
            .setStateBackend(defaultStateBackend)
            .setChangelogStateBackendEnabled(changelogStateBackendEnabled)
            .setSavepointDir(defaultSavepointDirectory)
            .setChaining(isChainingEnabled)
            .setUserArtifacts(cacheFile)
            .setTimeCharacteristic(timeCharacteristic)
            .setDefaultBufferTimeout(bufferTimeout)
            .setSlotSharingGroupResource(slotSharingGroupResources);
}

至此，我们终于找到了真正生成执行图的类 StreamGraphGenerator。这个我们稍后再说。
相信读者看到这里可能都忘了我们开始的地方，现在我们回到最初的 execute() 方法。

public JobExecutionResult execute(String jobName) throws Exception {
    final List<Transformation<?>> originalTransformations = new ArrayList<>(transformations);
    StreamGraph streamGraph = getStreamGraph();
    if (jobName != null) {
        streamGraph.setJobName(jobName);
    }

    try {
        return execute(streamGraph);
    } catch (Throwable t) {
        Optional<ClusterDatasetCorruptedException> clusterDatasetCorruptedException =
                ExceptionUtils.findThrowable(t, ClusterDatasetCorruptedException.class);
        if (!clusterDatasetCorruptedException.isPresent()) {
            throw t;
        }

        // Retry without cache if it is caused by corrupted cluster dataset.
        invalidateCacheTransformations(originalTransformations);
        streamGraph = getStreamGraph(originalTransformations);
        return execute(streamGraph);
    }
}

在完成了执行图的生成后，调用 execute(streamGraph)，将执行图赋给执行程序，并在出错后重新获取执行图再次执行。
接下来我们继续看 execute(streamGraph)

public JobExecutionResult execute(StreamGraph streamGraph) throws Exception {
    final JobClient jobClient = executeAsync(streamGraph);

    try {
        final JobExecutionResult jobExecutionResult;

        if (configuration.getBoolean(DeploymentOptions.ATTACHED)) {
            jobExecutionResult = jobClient.getJobExecutionResult().get();
        } else {
            jobExecutionResult = new DetachedJobExecutionResult(jobClient.getJobID());
        }

        jobListeners.forEach(
                jobListener -> jobListener.onJobExecuted(jobExecutionResult, null));

        return jobExecutionResult;
    } catch (Throwable t) {
        // get() on the JobExecutionResult Future will throw an ExecutionException. This
        // behaviour was largely not there in Flink versions before the PipelineExecutor
        // refactoring so we should strip that exception.
        Throwable strippedException = ExceptionUtils.stripExecutionException(t);

        jobListeners.forEach(
                jobListener -> {
                    jobListener.onJobExecuted(null, strippedException);
                });
        ExceptionUtils.rethrowException(strippedException);

        // never reached, only make javac happy
        return null;
    }
}

这个方法做了两件事：

• 调用真正执行的方法 executeAsync(streamGraph)返回 JobClient
• 针对执行结果，通过 jobClient.getJobExecutionResult().get() 获取

这里特别要提一下，JobClient 接口是任务执行的起点，负责接受用户的程序代码，然后创建数据流，将数据流提交给 JobManager 以便进一步执行。执行完成后，将结果返回给用户。这里就是通过 JobClient 取出执行结果 JobExecutionResult 对象。

不知道你是否注意到，任务完成后会执行 jobListeners 的 forEach 操作。jobListener 是 List 变量。
关于 JobListner 接口，源码注释如下：

/**
 * A listener that is notified on specific job status changed, which should be firstly registered by
 * {@code #registerJobListener} of execution environments.
 *
 * <p>It is highly recommended NOT to perform any blocking operation inside the callbacks. If you
 * block the thread the invoker of environment execute methods is possibly blocked.
 */
@PublicEvolving
public interface JobListener {

大意是在特定作业状态更改时被通知的侦听器，在 StreamExecutionEnvironment 中通过 registrJobListener 方法注册

public void registerJobListener(JobListener jobListener) {
    checkNotNull(jobListener, "JobListener cannot be null");
    jobListeners.add(jobListener);
}

而在任务执行完成后，会将其置为 null，表示执行 finished。
接下来我们继续看真正执行 execute 操作的 executeAsync(StreamGraph streamGraph) 方法

public JobClient executeAsync(StreamGraph streamGraph) throws Exception {
    checkNotNull(streamGraph, "StreamGraph cannot be null.");
    final PipelineExecutor executor = getPipelineExecutor();

    CompletableFuture<JobClient> jobClientFuture =
            executor.execute(streamGraph, configuration, userClassloader);

    try {
        JobClient jobClient = jobClientFuture.get();
        jobListeners.forEach(jobListener -> jobListener.onJobSubmitted(jobClient, null));
        collectIterators.forEach(iterator -> iterator.setJobClient(jobClient));
        collectIterators.clear();
        return jobClient;
    } catch (ExecutionException executionException) {
        final Throwable strippedException =
                ExceptionUtils.stripExecutionException(executionException);
        jobListeners.forEach(
                jobListener -> jobListener.onJobSubmitted(null, strippedException));

        throw new FlinkException(
                String.format("Failed to execute job '%s'.", streamGraph.getJobName()),
                strippedException);
    }
}

兜了一大圈，终于找到真正执行的方法，异步方法。
PipelineExecutor 按源码注释的解释是负责用户作业执行的实体，它由PipelineFactory 根据配置中确定的 Flink 环境按 yarn、standalone、per-job、local 几种不同情况生产对应的 Pipeline。这一点可以在 getPipelineExecutor 方法中得到证实

private PipelineExecutor getPipelineExecutor() throws Exception {
    checkNotNull(
            configuration.get(DeploymentOptions.TARGET),
            "No execution.target specified in your configuration file.");

    final PipelineExecutorFactory executorFactory =
            executorServiceLoader.getExecutorFactory(configuration);

    checkNotNull(
            executorFactory,
            "Cannot find compatible factory for specified execution.target (=%s)",
            configuration.get(DeploymentOptions.TARGET));

    return executorFactory.getExecutor(configuration);
}

获取对应环境的 PipelineExecutor 后调用接口中的 execute 方法执行，并将执行图、配置、类加载器作为参数传入
要想进一步搞清执行逻辑，我们必须继续深入探究 Pipeline.execute 的执行逻辑。
我们再官方文档中找到继承 Pipeline 接口的类有 AbstractJobClusterExecutor, AbstractSessionClusterExecutor, EmbeddedExecutor, KubernetesSessionClusterExecutor, LocalExecutor, RemoteExecutor, YarnJobClusterExecutor, YarnSessionClusterExecutor
下面我们就本地执行为例，探究 LocalExecutor 执行原理

public CompletableFuture<JobClient> execute(
            Pipeline pipeline, Configuration configuration, ClassLoader userCodeClassloader)
            throws Exception {
    checkNotNull(pipeline);
    checkNotNull(configuration);

    Configuration effectiveConfig = new Configuration();
    effectiveConfig.addAll(this.configuration);
    effectiveConfig.addAll(configuration);

    // we only support attached execution with the local executor.
    checkState(configuration.getBoolean(DeploymentOptions.ATTACHED));

    final JobGraph jobGraph = getJobGraph(pipeline, effectiveConfig);

    return PerJobMiniClusterFactory.createWithFactory(effectiveConfig, miniClusterFactory)
            .submitJob(jobGraph, userCodeClassloader);
}

该方法执行流程为：

• 将实例化 LocalExecutor 时添加的额外配置和用户配置合并为一个 Configuration
• 创建 JobGraph，作业执行图
• 调用 createWithFactory 和 submitJob 提交任务

这里的 JobGraph 是由 StreamGraph 转化而来，转化过程看 getJobGraph 方法
关于 JobGraph 的获取方法，我们留待后面讨论
这里 PerJobMiniClusterFactory.createWithFactory 创建了一个 PerJobMiniClusterFactory 对象
submitJob 开始了一个 MiniCluster 并提交了一个任务，具体代码如下：

public CompletableFuture<JobClient> submitJob(
            JobGraph jobGraph, ClassLoader userCodeClassloader) throws Exception {
    MiniClusterConfiguration miniClusterConfig =
            getMiniClusterConfig(jobGraph.getMaximumParallelism());
    MiniCluster miniCluster = miniClusterFactory.apply(miniClusterConfig);
    miniCluster.start();

    return miniCluster
            .submitJob(jobGraph)
            .thenApplyAsync(
                    FunctionUtils.uncheckedFunction(
                            submissionResult -> {
                                org.apache.flink.client.ClientUtils
                                        .waitUntilJobInitializationFinished(
                                                () ->
                                                        miniCluster
                                                                .getJobStatus(
                                                                        submissionResult
                                                                                .getJobID())
                                                                .get(),
                                                () ->
                                                        miniCluster
                                                                .requestJobResult(
                                                                        submissionResult
                                                                                .getJobID())
                                                                .get(),
                                                userCodeClassloader);
                                return submissionResult;
                            }))
            .thenApply(
                    result ->
                            new MiniClusterJobClient(
                                    result.getJobID(),
                                    miniCluster,
                                    userCodeClassloader,
                                    MiniClusterJobClient.JobFinalizationBehavior
                                            .SHUTDOWN_CLUSTER))
            .whenComplete(
                    (ignored, throwable) -> {
                        if (throwable != null) {
                            // We failed to create the JobClient and must shutdown to ensure
                            // cleanup.
                            shutDownCluster(miniCluster);
                        }
                    })
            .thenApply(Function.identity());
}

至此，我们总算明白了，execute 到最后是开启了一个 MiniCluster 并将 JobGraph 作为参数提交任务。
而 MiniCluster 在官方文档上的解释为 本地执行 Flink jobs 的 mini 集群。

总结：

• execute 的执行过程：
• 转换 Transformation 为 StreamGraph
• 提供执行需要的额外配置、监听方法等
• 将 StreamGraph 转换为可执行的 JobGraph
• 根据运行环境的不同创建不同的执行器
• 在本地环境下，开启一个 MiniCluster 并将 JobGraph 提交任务执行