Github user jackylk commented on the issue:
https://github.com/apache/carbondata/pull/2909
Since it is using CSV split in this PR, there are more tasks and each task
is smaller, thus we can get better parallelism and require less resource to run
in each task---
