PatrickRen commented on code in PR #19456:
URL: https://github.com/apache/flink/pull/19456#discussion_r850166745
##########
flink-connectors/flink-connector-kafka/src/main/java/org/apache/flink/connector/kafka/source/reader/KafkaPartitionSplitReader.java:
##########
@@ -98,9 +98,16 @@ public RecordsWithSplitIds<ConsumerRecord<byte[], byte[]>>
fetch() throws IOExce
ConsumerRecords<byte[], byte[]> consumerRecords;
try {
consumerRecords = consumer.poll(Duration.ofMillis(POLL_TIMEOUT));
- } catch (WakeupException we) {
- return new KafkaPartitionSplitRecords(
- ConsumerRecords.empty(), kafkaSourceReaderMetrics);
+ } catch (WakeupException | IllegalStateException e) {
+ // IllegalStateException will be thrown if the consumer is not
assigned any partitions.
+ // This happens if all assigned partitions are invalid or empty
(starting offset >=
Review Comment:
Thanks for the review! I think it's by-design that some kinds of starting /
stopping offsets like earliest, latest and committed are handled on
`SourceReader` (see `ReaderHandledOffsetsInitializer`) to avoid some race
conditions, like the earliest offset changes because of log deletion when split
is delivered from enumerator on JM to reader on TM.
--
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]
