[jira] [Created] (FLINK-10489) Inconsistent window information for streams with EventTime characteristic

["Alexis Sarda-Espinosa (JIRA)" <ji...@apache.org>]

Alexis Sarda-Espinosa created FLINK-10489:
---------------------------------------------

             Summary: Inconsistent window information for streams with EventTime characteristic
                 Key: FLINK-10489
                 URL: https://issues.apache.org/jira/browse/FLINK-10489
             Project: Flink
          Issue Type: Bug
         Environment: Flink v1.5.3
            Reporter: Alexis Sarda-Espinosa


Consider the following program.
{code:java}
void temp() throws Exception {
    StreamExecutionEnvironment execEnv = StreamExecutionEnvironment.getExecutionEnvironment();
    StreamTableEnvironment tableEnv = TableEnvironment.getTableEnvironment(execEnv);

    execEnv.setStreamTimeCharacteristic(TimeCharacteristic.EventTime);

    DataStream<Tuple3<Timestamp, String, Double>> source = execEnv
            .fromElements(
                    new Tuple3<>(Timestamp.valueOf("2018-09-20 22:00:00"), "a", 1.3),
                    new Tuple3<>(Timestamp.valueOf("2018-09-20 22:01:00"), "a", 2.1),
                    new Tuple3<>(Timestamp.valueOf("2018-09-20 22:02:00"), "a", 3.0),
                    new Tuple3<>(Timestamp.valueOf("2018-09-20 22:00:00"), "b", 2.2),
                    new Tuple3<>(Timestamp.valueOf("2018-09-20 22:01:00"), "b", 1.8)
            )
            .assignTimestampsAndWatermarks(new WaterMarker());

    Table table = tableEnv.fromDataStream(source, "f0, f1, f2, rowtime.rowtime")
            .window(Slide.over("2.minutes").every("1.minute").on("rowtime").as("w"))
            .groupBy("f1, w")
            .select("f1, w.end - 1.minute, f2.sum");

    tableEnv.toAppendStream(table, Row.class).print();
    execEnv.execute();
}

public static class WaterMarker implements AssignerWithPunctuatedWatermarks<Tuple3<Timestamp, String, Double>> {
    @Nullable
    @Override
    public Watermark checkAndGetNextWatermark(Tuple3<Timestamp, String, Double> lastElement, long extractedTimestamp) {
        return new Watermark(extractedTimestamp);
    }

    @Override
    public long extractTimestamp(Tuple3<Timestamp, String, Double> element, long previousElementTimestamp) {
        return element.f0.getTime();
    }
}
{code}
I am seeing an output like the following, where the hour is offset by 2 hours:
{code:java}
a,2018-09-20 20:00:00.0,1.3
a,2018-09-20 20:01:00.0,3.4000000000000004
a,2018-09-20 20:02:00.0,5.1
a,2018-09-20 20:03:00.0,3.0
{code}
My local time zone is GMT+01:00, and indeed the input timestamps are parsed with this time zone, but even if the window-end is printed as GMT, it should show 21:0x:00. In my tests, where I actually collect the output and can see the actual {{java.sql.Timestamp}} instances, I can always see this extra 1 hour offset.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)