[FLINK-18825] Support to process CDC message in batch mode

• Type: Sub-task
• Status: In Progress
• Priority: Major
• Resolution: Unresolved
• Component/s: Formats (JSON, Avro, Parquet, ORC, SequenceFile), Table SQL / Ecosystem
• Labels: auto-unassigned
• Assignee: Unassigned
• Reporter: Jark Wu
• Created: 05/Aug/20 07:50
• Updated: 20/Jul/21 02:35

Description

Currently, processing CDC is only supported for streaming mode. However, there is also cases to process CDC data in batch mode. For example, load, materialize and compress CDC data into Hive using parquet format.

Interpreting changelog in batch mode can be supported by adding a MaterializeOperator after the CDC source operator. The MaterializeOperator acts like a HashAggregate and SortAggregate based on whether all the fields are fixed-length. It will consume all the input RowDatas and replace/remove rows depending on the flag in the header.

Issue Links

• is related to: FLINK-23426 Support changelog processing in batch mode