Details
-
Sub-task
-
Status: Open
-
Major
-
Resolution: Unresolved
-
None
-
None
-
None
-
None
Description
In order to allow for more parallelism and predictable task sizes we should support Kudu key range splitting to allow more parallel tasks per tablet. Without this the parallelism is limited by the number of tablets to scan.
The implementation is like similar to the Spark implementation here:
https://github.com/apache/kudu/commit/22a6faa44364dec3a171ec79c15b814ad9277d8f