flink-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ever" <439674...@qq.com>
Subject 回复: [flink 1.8.1]window closed unexpectedly and data drop
Date Sun, 14 Jul 2019 09:03:36 GMT
timestamp of the fourth data is (03:17:55), and the watermark time should be 03:17:50(water
mark is 5). 
That time, window of the first data(ts:03:15:48) should be closed. 

What's more, there're so many sliding windows, and some of them should be closed to.

------------------ 原始邮件 ------------------
发件人: "Hequn Cheng"<chenghequn@gmail.com>;
发送时间: 2019年7月14日(星期天) 中午11:49
收件人: "Ever"<439674061@qq.com>;
抄送: "user"<user@flink.apache.org>;
主题: Re: [flink 1.8.1]window closed unexpectedly and data drop

Hi Ever,

The window only fires when the watermark passes the end of a window. 

> now the fourth data came with timestamp:03:17:55,   at that time, a new window should
be open, and the previous window should closed
The previous window may not close if the watermark hasn't passed the end of the window. More
info about watermark here[1]. 
Furthermore, we can monitor event time by checking the watermarks in the web dashboard[2].


[1] https://ci.apache.org/projects/flink/flink-docs-stable/dev/event_time.html#watermarks-in-parallel-streams
[2] https://ci.apache.org/projects/flink/flink-docs-master/monitoring/debugging_event_time.html#monitoring-current-event-time

On Sat, Jul 13, 2019 at 3:37 PM Ever <439674061@qq.com> wrote:

I have a streaming job based on Event time,  which has a 60 seconds window and 10 seconds
sliding window.
Data will come in batches every 10 second.

Here's the code.
val env = StreamExecutionEnvironment.getExecutionEnvironment


      .map(json => {
          new InvokingInfoWrapper(xxx)
      .assignTimestampsAndWatermarks(new BoundedOutOfOrdernessTimestampExtractor[InvokingInfoWrapper](Time.seconds(5))
        override def extractTimestamp(invoking: InvokingInfoWrapper): Long = {
      .keyBy(invokingInfo => {
      .timeWindow(Time.seconds(60), Time.seconds(10))
      .reduce(innerReducer).map(invokingInfo => {
 // ##2map =================================================
      //some mapping code
      .addSink(new WebSocketSink[InvokingInfoWrapper](wsHost)).name("Pangolin-websocket-sink")


And I Have noticed that something wrong:
1. the first data came with timestamp:03:15:48
2. the second data came with timestamp:03:15:59, and triggered the reduce operation(5 reduce
operations,there should be 5 window)
3. the third one: 03:16:06,   and also triggered reduce opertaions.
4. now the fourth data came with timestamp:03:17:55,   at that time, a new window should be
open, and the previous window should closed and the result should enter line "##2map" above.
But it didn't.
5. the fifth data came with timestamp:03:18:01, and triggered the reduce operation with the
fourth data.

So it seems that the top three datas had drop silently.

Somebody help on this?
View raw message