Github user xuchuanyin commented on the issue:
https://github.com/apache/carbondata/pull/3056
> because both the query and load flow were assigned the same taskId and
once query finished it freed the unsafe memory while the insert still in
progress.
How do you handle the scenario for stored-by-carbondata carbontable? In
that scenario, both of the query flow and load flow use offheap and encounter
the same problem just as you described above.
But I remembered we handle that differently from the current PR, which I
think their modification can be similar. Please check this again.
---
