I guess then word node in "Carbodata launches one job per each node to sort the data at node level and avoid shuffling" may make some confuse. I guess carbondata should launches one task per each executor . here job should be task ,node should be executor.
Maybe he can try increase the number of executors to avoid memory problem
