Details
-
Task
-
Status: Closed
-
Blocker
-
Resolution: Fixed
-
None
-
None
Description
We need to redesign the HoodieRecordPayload interface, which should not depend on avro, to pave the road for Spark Row writer work. Ideally, the new abstraction should have the individual implementation of avro-based record and spark-based Row, and the abstraction itself has more high-level operations like preCombine, getValue, etc., which are not tied to IndexedRecord. This should also make the write client easier to extend for Spark Dataset of Rows.
Attachments
Issue Links
- duplicates
-
HUDI-3217 RFC-46: Optimize Record Payload handling
- In Progress