"Buffer pool is destroyed"当我使用 Flink SlidingEventTimeWindows 时

Question

当我使用"SlidingEventTimeWindows"时，Flink抛出"java.lang.IollegalStateException：缓冲池被破坏"，但当我更改为"SlidingProcessingTimeWindows"后，一切都很好。

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$1.onProcessingTime(StreamSource.java:151)
	at org.apache.flink.streaming.runtime.tasks.SystemProcessingTimeService$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.access$301(ScheduledThreadPoolExecutor.java:180)
	at java.util.concurrent.ScheduledThreadPoolExecutor$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

我终于解决了以下步骤。

首先，将My DataMockSource中用于生成流数据的"collect"替换为"collectWithTimestamp"。这样做之后，控制台中将显示"发出延迟标记时出错"。

其次，将BoundedOutOfOrdernessTimestampExtractor替换为用于EventTime处理的AscendingTimestampExtractor。在我的DataMockSource中，我同时生成数据和发射，所以AscendingTimestampExtractor是生成水印的正确方法。

我在这里发布了主要代码，并在github上发布了完整的项目。希望它是有益的。

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

在事件时工作时，您需要安排在源中或使用assignTimestampsAndWatermarks进行时间戳提取和水印。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(事件时间窗口永远不会被触发(。

此外，您的源应该有一个cancel方法。类似这样的东西：

private volatile boolean running = true;
@Override
public void run(SourceContext ctx) throws Exception {
    while (running) {
        ...
    }
}
@Override
public void cancel() {
    running = false;
}

我想这也许可以解释你所看到的例外情况。在作业开始自行关闭后，源可能会继续运行并发送延迟标记。