"Buffer pool is destroyed" 当我使用 Flink SlidingEventTimeWindows

Question

Flink 在我使用 "SlidingEventTimeWindows" 时抛出 "java.lang.IllegalStateException: Buffer pool is destroyed"，但是当我更改为 "SlidingProcessingTimeWindows".

时一切正常

堆栈跟踪如下：

18:37:53,728 WARN  org.apache.flink.streaming.api.operators.AbstractStreamOperator  - Error while emitting latency marker.
java.lang.RuntimeException: Buffer pool is destroyed.
 at org.apache.flink.streaming.runtime.io.RecordWriterOutput.emitLatencyMarker(RecordWriterOutput.java:147)
 at org.apache.flink.streaming.api.operators.AbstractStreamOperator$CountingOutput.emitLatencyMarker(AbstractStreamOperator.java:683)
 at org.apache.flink.streaming.api.operators.StreamSource$LatencyMarksEmitter.onProcessingTime(StreamSource.java:151)
 at org.apache.flink.streaming.runtime.tasks.SystemProcessingTimeService$RepeatedTriggerTask.run(SystemProcessingTimeService.java:330)
 at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
 at java.util.concurrent.FutureTask.runAndReset(FutureTask.java:308)
 at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access1(ScheduledThreadPoolExecutor.java:180)
 at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:294)
 at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
 at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
 at java.lang.Thread.run(Thread.java:748)
Caused by: java.lang.IllegalStateException: Buffer pool is destroyed.
 at org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:230)
 at org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:204)
 at org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
 at org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
 at org.apache.flink.runtime.io.network.api.writer.RecordWriter.randomEmit(RecordWriter.java:125)
 at org.apache.flink.streaming.runtime.io.StreamRecordWriter.randomEmit(StreamRecordWriter.java:93)
 at org.apache.flink.streaming.runtime.io.RecordWriterOutput.emitLatencyMarker(RecordWriterOutput.java:144)
 ... 10 more

我终于解决了下面的步骤。

首先，将My DataMockSource中的"collect"替换为"collectWithTimestamp"，用于生成流data.After这样做，"Error while emitting latency marker"将在控制台中消失。

其次，将BoundedOutOfOrdernessTimestampExtractor替换为AscendingTimestampExtractor，用于EventTime processing.In我的DataMockSource，我生成数据并同时发出，所以AscendingTimestampExtractor是生成水印的正确方法。

我post这里的主要代码，以及github上的完整项目。希望对你有帮助。

StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
env.setStreamTimeCharacteristic(TimeCharacteristic.EventTime);
env.enableCheckpointing(10000); //

DataStreamSource<MockData> mockDataDataStreamSource = env.addSource(new DataMockSource());
mockDataDataStreamSource.assignTimestampsAndWatermarks(
    new AscendingTimestampExtractor<MockData>() {
      @Override
      public long extractAscendingTimestamp(MockData element) {
        return element.getTimestamp();
      }
    });

SingleOutputStreamOperator<Tuple2<String, Long>> countStream = mockDataDataStreamSource
    .keyBy("country").window(
        SlidingEventTimeWindows.of(Time.seconds(10), Time.seconds(10)))
//        .allowedLateness(Time.seconds(5))
    .process(
        new FlinkEventTimeCountFunction()).name("count elements");

countStream.addSink(new SinkFunction<Tuple2<String, Long>>() {
  @Override
  public void invoke(Tuple2<String, Long> value, Context context) throws Exception {
    System.out.println(value);
  }
});

env.execute("count test ");

我的 DataMockSource 在这里：

private volatile boolean running = true;
  @Override
  public void run(SourceContext sourceContext) throws Exception {
    while (running){
      MockData mockData = new MockData();
      mockData.setAge(ThreadLocalRandom.current().nextInt(1,99));
      mockData.setCountry("country "+ThreadLocalRandom.current().nextInt(2,5));
      mockData.setId(ThreadLocalRandom.current().nextLong());
      mockData.setTimestamp(Instant.now().toEpochMilli());
      // emit record with timestamp
      sourceContext.collectWithTimestamp(mockData,Instant.now().toEpochMilli());
//      sourceContext.collect(mockData);

      TimeUnit.SECONDS.sleep(3);
    }
  }

  @Override
  public void cancel() {
     running = false;
  }

Answer 1

在事件时间工作时，您需要在源中或使用 assignTimestampsAndWatermarks 安排时间戳提取和水印。看起来你没有这样做，这可以解释为什么你不会得到任何输出（事件时间 window 永远不会被触发）。

此外，您的来源应该有一个取消方法。像这样：

private volatile boolean running = true;

@Override
public void run(SourceContext ctx) throws Exception {
    while (running) {
        ...
    }
}

@Override
public void cancel() {
    running = false;
}

我认为这可以解释您所看到的异常。也许源继续运行并在作业开始自行关闭后发送延迟标记。

"Buffer pool is destroyed" 当我使用 Flink SlidingEventTimeWindows

"Buffer pool is destroyed" when I use Flink SlidingEventTimeWindows

apache-flink

flink-streaming