我的flink管道在下面看起来像
FlinkKafkaConsumerBase kafkaConsumer = new FlinkKafkaConsumer<>(topic, new DeserializationSchema(),props);
kafkaSource = env.addSource(kafkaConsumer).filter(<>);
WatermarkStrategy<GenericMetricV2> watermarkStrategy = WatermarkStrategy
.<GenericMetricV2>forBoundedOutOfOrderness(Duration.ofSeconds(900))
.withTimestampAssigner((metric, timestamp) -> {
logger.info("ETS: mts: {}, ts: {}", metric.metricPoint.timeInstant, timestamp);
return metric.metricPoint.timeInstant;
});
metricStream = kafkasource
.process(<>)
.assignTimestampsAndWatermarks(watermarkStrategy)
.transform("debugFilter", TypeInformation.of(<>), new StreamWatermarkDebugFilter<>("Op"))
.filter(<>)
.map(<>)
.flatMap(<>)
.keyBy(<>)
.window(TumblingEventTimeWindows.of(Time.seconds(300)))
.allowedLateneess(Time.seconds(900))
.sideOutputLateData(lateOutputTag)
.aggregate(AggregateFunction, ProcessWindowFunction)
.addSink()
我运行的并行度为1，默认setAutowatermarkInterval为200毫秒。我没有将setStreamTimeCharacteristic设置为flink 1.12，默认情况下它是事件时间。
我看到StreamWatermarkDebugFilter的输出正在进行水印处理，但所有事件都被标记为延迟，并在lateOutputTag处收集。
2021-05-18 17:14:19,745 INFO - ETS: mts: 1621310100000, ts: 1621310582271
2021-05-18 17:14:19,745 INFO - ETS: mts: 1621310100000, ts: 1621310582271
2021-05-18 17:14:19,842 INFO StreamWatermarkDebugFilter - Op, Watermark: 1621309499999
2021-05-18 17:14:19,944 INFO - ETS: mts: 1621309800000, ts: 1621310582275
2021-05-18 17:14:19,944 INFO - ETS: mts: 1621309800000, ts: 1621310582275
...
2021-05-18 17:14:20,107 INFO - ETS: mts: 1621310380000, ts: 1621310582278
2021-05-18 17:14:20,107 INFO - ETS: mts: 1621310380000, ts: 1621310582278
2021-05-18 17:14:20,137 INFO StreamWatermarkDebugFilter - Op, Watermark: 1621309779999
2021-05-18 17:14:20,203 INFO - ETS: mts: 1621309800000, ts: 1621310582279
...
2021-05-18 17:17:47,839 INFO - ETS: mts: 1621310100000, ts: 1621310681159
2021-05-18 17:17:47,848 INFO StreamWatermarkDebugFilter - Op, Watermark: 1621310099999
2021-05-18 17:17:47,958 INFO - ETS: mts: 1621309800000, ts: 1621310681237
2021-05-18 17:17:47,958 INFO - ETS: mts: 1621309800000, ts: 1621310681237
...
2021-05-18 17:22:24,207 INFO - ETS: mts: 1621310100000, ts: 1621310703622
2021-05-18 17:22:24,229 INFO StreamWatermarkDebugFilter - Op, Watermark: 1621310399999
2021-05-18 17:22:24,315 INFO - ETS: mts: 1621309800000, ts: 1621310705177
2021-05-18 17:22:24,315 INFO - ETS: mts: 1621309800000, ts: 1621310705177
我看到了这个讨论，它不是一个无所事事的问题。
这似乎与本次讨论有关。有人能建议我如何进一步调试这个问题，以确定可能是什么问题吗？
这是我没有分享的代码部分的问题。我在assignTimestampsAndWatermarks()之后做filter()，所以我不感兴趣的倾斜数据正在向前推水印。我把filter()移到了assignTimestampsAndWatermarks之前，它正在按预期工作。