Github user kl0u commented on a diff in the pull request: https://github.com/apache/flink/pull/3644#discussion_r108958127 --- Diff: flink-libraries/flink-cep/src/main/java/org/apache/flink/cep/operator/AbstractKeyedCEPPatternOperator.java --- @@ -196,19 +213,45 @@ public void processElement(StreamRecord<IN> element) throws Exception { updateNFA(nfa); } else { - getInternalWatermarkCallbackService().registerKeyForWatermarkCallback(keySelector.getKey(element.getValue())); - PriorityQueue<StreamRecord<IN>> priorityQueue = getPriorityQueue(); + // In event-time processing we assume correctness of the watermark. + // Events with timestamp smaller than the last seen watermark are considered late. + // Late events are put in a dedicated side output, if the user has specified one. + + if (element.getTimestamp() >= lastWatermark) { - // event time processing - // we have to buffer the elements until we receive the proper watermark - if (getExecutionConfig().isObjectReuseEnabled()) { - // copy the StreamRecord so that it cannot be changed - priorityQueue.offer(new StreamRecord<IN>(inputSerializer.copy(element.getValue()), element.getTimestamp())); + // we have an event with a valid timestamp, so + // we buffer it until we receive the proper watermark. + + getInternalWatermarkCallbackService().registerKeyForWatermarkCallback(keySelector.getKey(element.getValue())); + + PriorityQueue<StreamRecord<IN>> priorityQueue = getPriorityQueue(); + if (getExecutionConfig().isObjectReuseEnabled()) { + // copy the StreamRecord so that it cannot be changed + priorityQueue.offer(new StreamRecord<>(inputSerializer.copy(element.getValue()), element.getTimestamp())); + } else { + priorityQueue.offer(element); + } + updatePriorityQueue(priorityQueue); } else { - priorityQueue.offer(element); + sideOutputLateElement(element); } - updatePriorityQueue(priorityQueue); + } + } + + private void updateLastSeenWatermark(Watermark watermark) { + this.lastWatermark = watermark.getTimestamp(); --- End diff -- I believe that the `Watermark` handling should be the responsibility of the sources. Checkpointing it here would mean special logic for handling it in case of scaling down for example. In addition, when scaling down e.g from 2 tasks to 1 and the 2 tasks had different last watermarks, then we still may have issues with elements being processed out-of-order.
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. ---