[FLINK-11164] Refreshing expired shard iterator fails if no records were retrieved from shard since ShardConsumer was started in FlinkKinesisConsumer

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.8.0
• Component/s: Connectors / Kinesis
• Labels: pull-request-available
• Assignee: Aljoscha Krettek
• Reporter: Tzu-Li (Gordon) Tai
• Created: 14/Dec/18 08:26
• Updated: 04/Feb/19 10:38
• Resolved: 04/Feb/19 10:38

Description

Originally reported by: http://apache-flink-user-mailing-list-archive.2336050.n4.nabble.com/Encountered-the-following-Expired-Iterator-exception-in-getRecords-using-FlinkKinesisConsumer-td25093.html

In ShardConsumer.getRecords(...), we catch any expired iterator exceptions so that in the case no data was written to the Kinesis shard, we last iterator which eventually expires doesn't just fail the job, by doing:

shardItr = kinesis.getShardIterator(subscribedShard, ShardIteratorType.AFTER_SEQUENCE_NUMBER.toString(), lastSequenceNum.getSequenceNumber());

The problem is that, if absolutely no records were retrieved at all from the shard since the ShardConsumer was started, then lastSequenceNumber would still be a sentinel value (e.g. EARLIEST_SEQUENCE_NUMBER, LATEST_SEQUENCE_NUMBER, etc.) instead of an actual value. This isn't recognized by Kinesis, and would fail the job.

Issue Links

• links to: GitHub Pull Request #7601