Details
-
New Feature
-
Status: Open
-
Major
-
Resolution: Unresolved
-
v4.0.0-beta
-
None
Description
We'll always need to filter data using partition column, but no need to include partition in aggregation dimensions. For these cases, Kylin will choose cuboid that include partition column. It's not efficient enough for those queries and we can't take the best advantage of precomputing.
Recently we have tried to split a query into two parts, one part of query will process segments whose data is fully used in a query so that it can choose cuboid without partition column. This dose bring great improvement for big queries, one of our test cases' response time reduced from 20s to 6s. We'll refine the code and share it.