背景

在Apache Hudi初探(一)(与flink的结合)中，我们提到了Pipelines.hoodieStreamWrite 写hudi文件,这个操作真正写hudi是在Pipelines.hoodieStreamWrite方法下的transform(opName("stream_write", conf), TypeInformation.of(Object.class), operatorFactory),具体分析一下写入的过程。

分析

对于transform(opName("stream_write", conf), TypeInformation.of(Object.class), operatorFactory)这个代码片段，我们主要看operatorFactory 这个对象(transform这个操作是Flink框架的操作):

public class StreamWriteOperator<I> extends AbstractWriteOperator<I> {

  public StreamWriteOperator(Configuration conf) {
    super(new StreamWriteFunction<>(conf));
  }

  public static <I> WriteOperatorFactory<I> getFactory(Configuration conf) {
    return WriteOperatorFactory.instance(conf, new StreamWriteOperator<>(conf));
  }
}

最主要的hudi算子为StreamWriteOperator,其中最主要的操作是由StreamWriteFunction来完成的：

// StreamWriteFunction
  

   @Override
  public void initializeState(FunctionInitializationContext context) throws Exception {
    this.taskID = getRuntimeContext().getIndexOfThisSubtask();
    this.metaClient = StreamerUtil.createMetaClient(this.config);
    this.writeClient = FlinkWriteClients.createWriteClient(this.config, getRuntimeContext());
    this.writeStatuses = new ArrayList<>();
    this.writeMetadataState = context.getOperatorStateStore().getListState(
        new ListStateDescriptor<>(
            "write-metadata-state",
            TypeInformation.of(WriteMetadataEvent.class)
        ));

    this.ckpMetadata = CkpMetadata.getInstance(this.metaClient.getFs(), this.metaClient.getBasePath());
    this.currentInstant = lastPendingInstant();
    if (context.isRestored()) {
      restoreWriteMetadata();
    } else {
      sendBootstrapEvent();
    }
    // blocks flushing until the coordinator starts a new instant
    this.confirming = true;
  }

  @Override
  public void open(Configuration parameters) throws IOException {
    this.tracer = new TotalSizeTracer(this.config);
    initBuffer();
    initWriteFunction();
  }

  @Override
  public void snapshotState(FunctionSnapshotContext functionSnapshotContext) throws Exception {
    if (inputEnded) {
      return;
    }
    snapshotState();
    // Reload the snapshot state as the current state.
    reloadWriteMetaState();
  }

  @Override
  public void snapshotState() {
    // Based on the fact that the coordinator starts the checkpoint first,
    // it would check the validity.
    // wait for the buffer data flush out and request a new instant
    flushRemaining(false);
  }

  @Override
  public void processElement(I value, ProcessFunction<I, Object>.Context ctx, Collector<Object> out) throws Exception {
    bufferRecord((HoodieRecord<?>) value);
  }

• initializeState操作，主要是做一些初始化的操作

  • this.taskID = getRuntimeContext().getIndexOfThisSubtask();
    获取当前的task的索引下标，用来向operator coordinator发送event给operator coordinator,之后 StreamWriteOperatorCoordinator(operator coordinator) 进行处理，后续会说到StreamWriteOperatorCoordinator

  • metaClient = StreamerUtil.createMetaClient(this.config)
writeClient = FlinkWriteClients.createWriteClient
    初始化hudi的元数据客户端(这里是HoodieTableMetaClient)和写入客户端(这里是HoodieFlinkWriteClient)

  • writeStatuses = new ArrayList<>()
    记录后续的写入hudi文件的信息

  • writeMetadataState = context.getOperatorStateStore().getListState
    记录写入hudi的元数据事件，会在后续的操作中，会包装成event发送给operator coordinator(StreamWriteOperatorCoordinator)

  • ckpMetadata = CkpMetadata.getInstance
    Flink的checkpoint的元数据信息路径,默认的路径是/${hoodie.basePath}/.hoodie/.aux/ckp_meta

  • currentInstant = lastPendingInstant()
    获取上次还没有完成的commit

  • restoreWriteMetadata或者sendBootstrapEvent，根据是否是从checkpoint恢复过来的进行不同消息的发送，
    这里的operator coordinator(StreamWriteOperatorCoordinator)会进行统一的处理,并初始化一个commit

• open操作
  写入hudi前的前置操作，比如说 初始化TotalSizeTracer记录maxBufferSize便于flush操作
  根据write.operation的值（默认是upsert）选择后续的操作是insert或upsert或overwrite,这里是upsert

• processElement操作
  这里对传入的HoodieRecord进行缓存，主要是bufferRecord做的事情，

  • 首先会获取bucketID，之后再往对应的bucket中插入数据
  • 如果超出write.batch.size(默认是128MB)，则会进行flushBucket操作，该操作主要是写入hudi操作 //TODO： 具体的写入hudi操作
    • 首先会获取新的需要提交的commit
    • 再进行写入的实际操作
    • 写入的文件元数据信息回传到operator coordinator进行统一处理

• snapshotState 操作

  • 调用flushRemaining 写入剩下的数据到hudi存储中
  • 重新加载当前写入的hudi文件元数据信息到当前flink的state中

hudi StreamWriteOperatorCoordinator作用

总的来说，StreamWriteOperatorCoordinator扮演的角色和在Spark中driver的角色一样，都是来最后来提交 元数据信息到huid中。
具体的作用还是得从具体的方法来看：

  @Override
  public void handleEventFromOperator(int i, OperatorEvent operatorEvent) {
    ValidationUtils.checkState(operatorEvent instanceof WriteMetadataEvent,
        "The coordinator can only handle WriteMetaEvent");
    WriteMetadataEvent event = (WriteMetadataEvent) operatorEvent;

    if (event.isEndInput()) {
      // handle end input event synchronously
      // wrap handleEndInputEvent in executeSync to preserve the order of events
      executor.executeSync(() -> handleEndInputEvent(event), "handle end input event for instant %s", this.instant);
    } else {
      executor.execute(
          () -> {
            if (event.isBootstrap()) {
              handleBootstrapEvent(event);
            } else {
              handleWriteMetaEvent(event);
            }
          }, "handle write metadata event for instant %s", this.instant
      );
    }
  }
  ...
  @Override
  public void notifyCheckpointComplete(long checkpointId) {
    executor.execute(
        () -> {
          // The executor thread inherits the classloader of the #notifyCheckpointComplete
          // caller, which is a AppClassLoader.
          Thread.currentThread().setContextClassLoader(getClass().getClassLoader());
          // for streaming mode, commits the ever received events anyway,
          // the stream write task snapshot and flush the data buffer synchronously in sequence,
          // so a successful checkpoint subsumes the old one(follows the checkpoint subsuming contract)
          final boolean committed = commitInstant(this.instant, checkpointId);

          if (tableState.scheduleCompaction) {
            // if async compaction is on, schedule the compaction
            CompactionUtil.scheduleCompaction(metaClient, writeClient, tableState.isDeltaTimeCompaction, committed);
          }

          if (tableState.scheduleClustering) {
            // if async clustering is on, schedule the clustering
            ClusteringUtil.scheduleClustering(conf, writeClient, committed);
          }

          if (committed) {
            // start new instant.
            startInstant();
            // sync Hive if is enabled
            syncHiveAsync();
          }
        }, "commits the instant %s", this.instant
    );
  }

• handleEventFromOperator方法用来接受task发送的消息

  • 对于BootStrap类型的WriteMetadataEvent(在StreamWriteFunction方法initializeState中),相当于函数初始化也就会触发
    该类型的消息由handleBootstrapEvent来处理(我们这里假设每个任务operator都完成了初始化的操作)，对应的数据流如下：

    initInstant
       ||
       \/
    reset => startInstant
    

    startInstant 这里就会初始化一个hudi写操作的commit信息

  • 对于一般的write的信息的event，(比如说在processElement的flushBucket函数中)，由handleWriteMetaEvent来处理：

     if (this.eventBuffer[event.getTaskID()] != null) {
     this.eventBuffer[event.getTaskID()].mergeWith(event);
     } else {
       this.eventBuffer[event.getTaskID()] = event;
     }
    

    这里只是加到变量名为eventBuffer 的WriteMetadataEvent类型的数组中，后续中会进行处理

  • 对于isEndInput为true的event，这种一般source是基于文件的这种，这里先不讨论

• notifyCheckpointComplete 当对应的checkpointId完成以后，该方法会被调用

  • commitInstant 提交hudi元数据，如果如果有发生异常，则回滚当前hudi对应的commit
  • scheduleCompaction && scheduleClustering 进行hui的Compcation和Clustering
  • 如果成功的提交了,则会开启一个新的commit，如果开了hive同步(hive_sync.enabled默认为false)，则会同步元数据信息到hive

总结

用一张图总结一下交互方式，如下：