Uploaded image for project: 'Flink'
  1. Flink
  2. FLINK-35115

Kinesis connector writes wrong Kinesis sequence number at stop with savepoint

    XMLWordPrintableJSON

Details

    • Bug
    • Status: Resolved
    • Blocker
    • Resolution: Fixed
    • 1.15.4, aws-connector-4.0.0, aws-connector-4.1.0, aws-connector-4.2.0, 1.16.3, 1.17.2, 1.18.1
    • aws-connector-4.3.0
    • Connectors / Kinesis
    • Important

    Description

      Having an exactly-once Kinesis -> Flink -> Kafka job and triggering a stop-with-savepoint, Flink duplicates in Kafka all the records between the last checkpoint and the savepoint at resume:

      • Event1 is written to Kinesis
      • Event1 is processed by Flink 
      • Event1 is committed to Kafka at the checkpoint
      • ............................................................................
      • Event2 is written to Kinesis
      • Event2 is processed by Flink
      • Stop with savepoint is triggered manually
      • Event2 is committed to Kafka
      • ............................................................................
      • Job is resumed from the savepoint
      • Event2 is written again to Kafka at the first checkpoint

       

      I believe that it's a Kinesis connector issue for 2 reasons:

      • I've checked the actual Kinesis sequence number in the _metadata file generated at stop-with-savepoint and it's the one from the checkpoint before the savepoint  instead of being the one of the last record committed to Kafka.
      • I've tested exactly the save job with Kafka as source instead of Kinesis as source and the behaviour does not reproduce.

      Attachments

        Issue Links

          Activity

            People

              a.pilipenko Aleksandr Pilipenko
              vadim.vararu Vadim Vararu
              Votes:
              0 Vote for this issue
              Watchers:
              4 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved: