从Beam/数据流读取RabbitMQ

我正在尝试以流方式从beam/dataflow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();
    }
}

如果我不将withMaxReadTime传递给RabbitMqIO。 如果我传入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

---

我如何设置一个流式流，使其无限期地从RabbitMQ运行？

我对数据流管道也有类似的问题。尝试在数据流中运行时，我得到：

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)

问题是RabbitMqIO使用来自RabbitMq的消息的时间戳，例如用于水印。事实证明，在我的例子中，来自RabbitMq的消息没有设置时间戳。RabbitMq中默认情况下没有设置时间戳，它为null。我通过为ApacheBeam中的类准备补丁修复了这个问题。我在RabbitMqMessage构造函数中进行了更改。现在看起来是这样的：

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();
}

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();
    }
}

我必须更改RabbitMqIO中的advance方法，以不使用可能为null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;
    }

在再次运行管道后，我在另一个地方再次遇到此异常。这一次是由于没有在RabbitMQCheckpointMark中为OldTestTimeStamp属性设置默认值造成的。我做了下一个更改，现在RabbitMQCheckpointMark看起来如下：

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();
}

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();
    }
}

---

所有这些更改修复了我的管道，现在它按预期工作。我希望您会发现它很有用。

这是Io中的一个错误，已经被删除