Admaing opened a new pull request, #29331: URL: https://github.com/apache/flink/pull/29331
Backport of https://github.com/apache/flink/pull/29239 (91e1e5fba04a .. 0897586aa505), which went to master but not to this branch. A FLIP-27 source that was announced as IDLE downstream stayed IDLE after it resumed, unless its next watermark strictly advanced the combined watermark: `WatermarkOutputMultiplexer` never reported the active state of the combined status, `WatermarksWithIdleness` never undid the `markIdle()` it had emitted, and `WatermarkToDataOutput` marked the output active only after the monotonicity guard. Records of a resumed source could therefore be dropped as late events. The four commits are cherry-picked without conflicts; the touched files are identical on `release-2.3`, and the new test file is the only addition. Verified locally on JDK 17: `*Watermark*Test`, `SourceOperatorEventTimeTest` and `SourceOperatorAlignmentTest` pass (22 test classes, 158 tests). That includes the new `ProgressiveTimestampsAndWatermarksSubtaskIdlenessTest` and the `SourceOperatorAlignmentTest` case that an earlier version of the fix broke. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
