Github user NamanRastogi commented on the issue:
https://github.com/apache/carbondata/pull/2850
@xuchuanyin
Using this API the overhead of reading concurrently is on the consumer of
CarbonReader. So yes, what you said what you said is right. But this API was
made to consider this case only when a user wants to have more manual control
over the concurrent reading. Consider the scenario when the user wants to read
different files in different machines. If we take care of the concurrent
reading internally (using a buffer as you suggested), the user cannot do that.
Another thing that you mentioned was user providing the location of
_carbondata_ files in the SDK reader itself, and that does not work in the
current implementation of SDK.
And for faster reading, there is one more pull request:
[2816](https://github.com/apache/carbondata/pull/2816), it supports batch
reading for a better reading performance. It is still not concurrent reading,
but better than iterator based row-by-row reading.
---
