Details
-
Improvement
-
Status: Closed
-
Major
-
Resolution: Abandoned
-
None
-
None
-
None
Description
We're now serializing record into a ByteBuffer and store it to a single field.
That was needed for Kafka and Redis, but I feel it's not natural for some data sources which supports schema (like JDBC, MongoDB, etc.)
For these data sources, record should be stored as it is (not serialized).
If new approach doesn't break backward compatibility it should be great, but if it's necessary to break, I think it's OK to break since Storm SQL is still marked as 'experiment'.