In 1.5.3 for
KYLIN-1677, Kylin changed the steps when pull data from hive as: 1) count the source table 2) create intermediate flat table with redistribution (reducer number is determined by step1)
This works good for many cases, but may not benefit for case like a view (which joins several tables) as the fact table.
So it is better to keep the other option: 1) create intermediate flat table 2) count the intermediate table 3) redistribute it based on the output of step 2)
Plan to make this configurable (system level & cube level).