Hi everyone, we’re experiencing a strange phenomenon when using the table API and converting a table to a data stream with `toDataStream()`: the watermarks that we had assigned in the

CREATE TABLE

statement seem to disappear and we account the following error:

java.lang.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(...)'?
	at org.apache.flink.streaming.api.windowing.assigners.SlidingEventTimeWindows.assignWindows(SlidingEventTimeWindows.java:80) ~[flink-streaming-java-1.16.1.jar:1.16.1]
	at org.apache.flink.streaming.runtime.operators.windowing.WindowOperator.processElement(WindowOperator.java:284) ~[flink-streaming-java-1.16.1.jar:1.16.1]
...

We are currently fixing it by assigning timestamps a second time, after the conversion, but feel like this shouldn’t be necessary. Are we misunderstanding how watermarks work (especially in regard to mixing Table and Datastream API) or did we possibly encounter a bug? Is reassigning watermarks a valid approach in general? To give more context, we’re using Flink 1.16.1 with the Java API and consume a Kinesis stream (although consuming from a file source yields the same behaviour). In the job we have two aggregations that use different

SlidingEventTimeWindows

and an

intervalJoin

to join the results.

I encountered the same issue as well. Any chance you have found out the root cause of this?

Unfortunately not. Although we haven’t encountered any issues so far by simply reapplying the watermarks, I am still curious about the underlying cause of this as well.

Nws, thanks for the tips here anyway as this has unblocked me 😃