话不多说,直接上手今天的主题,探索一个容易让人忽略和困惑的问题:Flink 时间窗口的起始时间
就以最简单的demo为例:
timeWindow(Time.seconds(5))
上述定义一个步长为5s的滚动窗口,就以这个简单的入口进入Flink的源码开始探索
1)timeWindow的定义
public WindowedStream<T, KEY, TimeWindow> timeWindow(Time size) {
if (environment.getStreamTimeCharacteristic() == TimeCharacteristic.ProcessingTime) {
return window(TumblingProcessingTimeWindows.of(size));
} else {
return window(TumblingEventTimeWindows.of(size));
}
}
这段源码比较贴近大众,就是一个普通的判断,而且environment.getStreamTimeCharacteristic()这个东西我们再熟悉不过了,判断当前是ProcessingTime还是EventTime,当然除了EventTime还有IngestionTime,但是比较常用的还是ProcessingTime和EventTime,所以我们就非ProcessingTime即EventTime这样理解,因为生产环境比较常用的是EventTime,所以我们就进入else的代码继续查看
2)TumblingEventTimeWindows的定义
window(TumblingEventTimeWindows.of(size))这段代码,window利用TumblingEventTimeWindows来分配元素,所以我们要了解的核心是TumblingEventTimeWindows.of(size)的定义
public static TumblingEventTimeWindows of(Time size) {
return new TumblingEventTimeWindows(size.toMilliseconds(), 0);
}
protected TumblingEventTimeWindows(long size, long offset) {
if (Math.abs(offset) >= size) {
throw new IllegalArgumentException
("TumblingEventTimeWindows parameters must satisfy abs(offset) < size");
}
this.size = size;
this.offset = offset;
}
可以看到通过of方法我们构建了一个offset为0,size为5的TumblingEventTimeWindows对象,然后就是我们需要的核心方法,assignWindows,窗口分配元素的核心方法
@Override
public Collection<TimeWindow> assignWindows(Object element, long timestamp, WindowAssignerContext context) {
if (timestamp > Long.MIN_VALUE) {
// Long.MIN_VALUE is currently assigned when no timestamp is present
long start = TimeWindow.getWindowStartWithOffset(timestamp, offset, size);
return Collections.singletonList(new TimeWindow(start, start + size));
} else {
throw new RuntimeException(
"Record has Long.MIN_VALUE timestamp (= no timestamp marker). " +
"Is the time characteristic set to 'ProcessingTime',
or did you forget to call " +
"'DataStream.assignTimestampsAndWatermarks(...)'?");
}
}
重点来了
long start = TimeWindow.getWindowStartWithOffset(timestamp, offset, size);
/**
* Method to get the window start for a timestamp.
*
* @param timestamp epoch millisecond to get the window start.
* @param offset The offset which window start would be shifted by.
* @param windowSize The size of the generated windows.
* @return window start
*/
public static long getWindowStartWithOffset(long timestamp, long offset, long windowSize) {
return timestamp - (timestamp - offset + windowSize) % windowSize;
}
Method to get the window start for a timestamp.翻译过来就是这个方法用来获取窗口的开始时间戳
核心算法就是
timestamp - (timestamp - offset + windowSize) % windowSize
上一段代码小测一下
case class TestData(timestamp:Long,word:String) def main(args: Array[String]): Unit = { val env = StreamExecutionEnvironment.getExecutionEnvironment env.setStreamTimeCharacteristic(TimeCharacteristic.EventTime) // 便于输出,设置并行度为1 env.setParallelism(1) val socketStream = env.socketTextStream("localhost",9999) val windowedStream = socketStream .map(row=>TestData(row.split(" ")(0).toLong,row.split(" ")(0))) .assignTimestampsAndWatermarks(new BoundedOutOfOrdernessTimestampExtractor[TestData](Time.seconds(1)) { override def extractTimestamp(element: TestData): Long = element.timestamp * 1000 }) .keyBy(_.word) .timeWindow(Time.seconds(5)) .reduce((r1,r2)=>TestData(r1.timestamp,"hello "+r2.word)) windowedStream.print("window output is") socketStream.print("input data is") env.execute("window_test_job") }
准备一下测试数据 1599623712 word(2020-09-09 11:55:12) 1599623715 word(2020-09-09 11:55:15)
根据公式算出开始时间: 1599623712 - (1599623712 - 0 + 5) % 5 == 1599623710 也就是开始时间为 1599623710,步长为5s,也就是下次触发窗口计算为1599623715
验证一下:
nc录入数据:1599623712 word 1599623715 word
控制台输出结果:
input data is> 1599623712 word input data is> 1599623715 word window output is> TestData(1599623712,word)
结果验证了公式结果即为窗口的开始时间,ProcessingTime与之类似就不测试了,其实也可以看到公式的计算结果一般为自然时间的开始,如2020-09-09 11:55:12的开始时间为2020-09-09 11:55:10