从 Beam/DataFlow 读取 RabbitMQ

Question

我正在尝试以流方式从光束/数据流运行 RabbitMQ 队列（以便它无限期地运行。）

我尝试运行的最小示例代码是：

import org.apache.beam.sdk.Pipeline;
import org.apache.beam.sdk.io.rabbitmq.RabbitMqIO;
import org.apache.beam.sdk.io.rabbitmq.RabbitMqMessage;
import org.apache.beam.sdk.transforms.DoFn;
import org.apache.beam.sdk.transforms.ParDo;

public class RabbitMqTest {

    public static void main(String[] args) {
        Pipeline pipeline = Pipeline.create();
        final String serverUri = "amqp://guest:guest@localhost:5672";

        pipeline
                .apply("Read RabbitMQ message", RabbitMqIO.read().withUri(serverUri).withQueue("my_queue"))
                .apply(ParDo.of(new DoFn<RabbitMqMessage, String>() {

                    @ProcessElement
                    public void processElement(ProcessContext c) {
                        String message = new String(c.element().getBody());
                        System.out.println();
                        c.output(message);
                    }
                }));
        pipeline.run().waitUntilFinish();
    }
}

但是它崩溃了：

Exception in thread "main" java.lang.NullPointerException
at org.apache.beam.runners.direct.UnboundedReadEvaluatorFactory$UnboundedReadEvaluator.processElement(UnboundedReadEvaluatorFactory.java:169)
at org.apache.beam.runners.direct.DirectTransformExecutor.processElements(DirectTransformExecutor.java:160)
at org.apache.beam.runners.direct.DirectTransformExecutor.run(DirectTransformExecutor.java:124)
at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
at java.util.concurrent.FutureTask.run(FutureTask.java:266)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
at java.lang.Thread.run(Thread.java:748)

如果我没有将 withMaxReadTime() 传递给 RabbitMqIO。 如果我确实传入了 withMaxReadTime()，它会阻塞 X 秒，然后处理在此期间到达的所有消息，然后退出。

如何设置无限期地从 RabbitMQ 运行的流式传输流？

Answer 1

Dataflow 管道也有类似的问题。当试图在 Dataflow 中运行它时，我得到了：

java.lang.NullPointerException
            org.apache.beam.runners.dataflow.worker.WindmillTimeUtils.harnessToWindmillTimestamp(WindmillTimeUtils.java:58)
            org.apache.beam.runners.dataflow.worker.StreamingModeExecutionContext.flushState(StreamingModeExecutionContext.java:400)
            org.apache.beam.runners.dataflow.worker.StreamingDataflowWorker.process(StreamingDataflowWorker.java:1230)
            org.apache.beam.runners.dataflow.worker.StreamingDataflowWorker.access$1000(StreamingDataflowWorker.java:143)
            org.apache.beam.runners.dataflow.worker.StreamingDataflowWorker$6.run(StreamingDataflowWorker.java:967)
            java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
            java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
            java.lang.Thread.run(Thread.java:745)

他们的问题是RabbitMqIO 使用来自 RabbitMq 的消息中的时间戳，例如用于水印。事实证明，在我的情况下，来自 RabbitMq 的消息没有设置时间戳（默认情况下在 RabbitMq 中未设置）并且它为空。我通过为 Apache Beam 中的类准备补丁来解决这个问题。我对RabbitMqMessage 构造函数进行了更改。现在看起来像这样：

public RabbitMqMessage(String routingKey, QueueingConsumer.Delivery delivery) {
    this.routingKey = routingKey;
    body = delivery.getBody();
    contentType = delivery.getProperties().getContentType();
    contentEncoding = delivery.getProperties().getContentEncoding();
    headers = delivery.getProperties().getHeaders();
    deliveryMode = delivery.getProperties().getDeliveryMode();
    priority = delivery.getProperties().getPriority();
    correlationId = delivery.getProperties().getCorrelationId();
    replyTo = delivery.getProperties().getReplyTo();
    expiration = delivery.getProperties().getExpiration();
    messageId = delivery.getProperties().getMessageId();
    /*
      *** IMPORTANT ***
      Sometimes timestamp in RabbitMq message properties is 'null'. `RabbitMqIO` uses that value as
      watermark, when it is `null` it causes exceptions, 'null' has to be replaced with some value in this case current time
     */
    // timestamp = delivery.getProperties().getTimestamp();
    timestamp = delivery.getProperties().getTimestamp() == null ? new Date() : delivery.getProperties().getTimestamp();
    type = delivery.getProperties().getType();
    userId = delivery.getProperties().getUserId();
    appId = delivery.getProperties().getAppId();
    clusterId = delivery.getProperties().getClusterId();
}

我不得不更改RabbitMqIO 中的advance() 方法以不使用可能为空的timestamp 属性：

@Override
    public boolean advance() throws IOException {
        try {
            QueueingConsumer.Delivery delivery = consumer.nextDelivery(1000);
            if (delivery == null) {
                return false;
            }
            if (source.spec.useCorrelationId()) {
                String correlationId = delivery.getProperties().getCorrelationId();
                if (correlationId == null) {
                    throw new IOException(
                            "RabbitMqIO.Read uses message correlation ID, but received "
                                    + "message has a null correlation ID");
                }
                currentRecordId = correlationId.getBytes(StandardCharsets.UTF_8);
            }
            long deliveryTag = delivery.getEnvelope().getDeliveryTag();
            checkpointMark.sessionIds.add(deliveryTag);

            current = new RabbitMqMessage(source.spec.routingKey(), delivery);
            /*
              *** IMPORTANT ***
              Sometimes timestamp in RabbitMq messages is 'null' stream in Dataflow fails because
              watermark is based on that value, 'null' has to be replaced with some value. `RabbitMqMessage` was changed
              to use `new Date()` in this situation and now timestamp can be taken from it
             */
            //currentTimestamp = new Instant(delivery.getProperties().getTimestamp());
            currentTimestamp = new Instant(current.getTimestamp());
            if (currentTimestamp.isBefore(checkpointMark.oldestTimestamp)) {
                checkpointMark.oldestTimestamp = currentTimestamp;
            }
        } catch (Exception e) {
            throw new IOException(e);
        }
        return true;
    }

再次运行我的管道后，我在其他地方再次遇到了这个异常。这次是因为没有在RabbitMQCheckpointMark 中设置oldestTimestamp 属性的默认值。我做了下一个更改，现在RabbitMQCheckpointMark 看起来像这样：

private static class RabbitMQCheckpointMark
        implements UnboundedSource.CheckpointMark, Serializable {
    transient Channel channel;
    /*
      *** IMPORTANT *** it should be initialized with some value because without it runner (e.g Dataflow) fails with 'NullPointerException'
      Example error:
        java.lang.NullPointerException
            org.apache.beam.runners.dataflow.worker.WindmillTimeUtils.harnessToWindmillTimestamp(WindmillTimeUtils.java:58)
            org.apache.beam.runners.dataflow.worker.StreamingModeExecutionContext.flushState(StreamingModeExecutionContext.java:400)
            org.apache.beam.runners.dataflow.worker.StreamingDataflowWorker.process(StreamingDataflowWorker.java:1230)
            org.apache.beam.runners.dataflow.worker.StreamingDataflowWorker.access$1000(StreamingDataflowWorker.java:143)
            org.apache.beam.runners.dataflow.worker.StreamingDataflowWorker$6.run(StreamingDataflowWorker.java:967)
            java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
            java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
            java.lang.Thread.run(Thread.java:745)
     */
    Instant oldestTimestamp = new Instant(Long.MIN_VALUE);
    final List<Long> sessionIds = new ArrayList<>();

    @Override
    public void finalizeCheckpoint() throws IOException {
        for (Long sessionId : sessionIds) {
            channel.basicAck(sessionId, false);
        }
        channel.txCommit();
        oldestTimestamp = Instant.now();
        sessionIds.clear();
    }
}

所有这些更改都修复了我的管道，现在它可以按预期工作。我希望你会发现它有用。

Answer 2

这是 Io 中的一个错误，一直是 fixed recently。