PatrickRen commented on code in PR #54:
URL:
https://github.com/apache/flink-connector-kafka/pull/54#discussion_r1351861716
##########
flink-connector-kafka/src/main/java/org/apache/flink/connector/kafka/source/reader/KafkaPartitionSplitReader.java:
##########
@@ -110,6 +110,13 @@ public RecordsWithSplitIds<ConsumerRecord<byte[], byte[]>>
fetch() throws IOExce
markEmptySplitsAsFinished(recordsBySplits);
return recordsBySplits;
}
+
+ // Track the record fetch lag
+ consumerRecords
+ .iterator()
+ .forEachRemaining(
+ record ->
kafkaSourceReaderMetrics.recordFetched(record.timestamp()));
Review Comment:
Actually it is not accurate to use `ConsumerRecord#timestamp` directly as
the event time. Users can specify their own `WatermarkStrategy` to extract
event time from the payload of the record instead of using the timestamp on
`ConsumerRecord`:
https://github.com/apache/flink/blob/6b52a4107db7521a25f4f308891095c5ba33cca0/flink-streaming-java/src/main/java/org/apache/flink/streaming/api/environment/StreamExecutionEnvironment.java#L2025
Fetch time is determined in split reader in fetcher manager thread, while
event time is determined in SourceOperator in task's main thread, so there's a
gap between, and that's why we didn't implement this metric in the first time.
--
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]
