On 1/26/15, 9:18 AM, Philippe Kernévez wrote:
This degradation is due to this bug (requests are replayed with a full scan) : https://issues.apache.org/jira/browse/HIVE-9382
I doubt that is the issue you are hitting, if you're moving from 0.13 to 0.14.
You are possibly hitting HIVE-9401. To confirm that, please set the fetch conversion to "more" and change the threshold parameters for the query
set hive.fetch.task.conversion.threshold=<size-of-table-in-bytes>;
I tried to change the value of "set hive.fetch.task.conversion=none", but it doesn't change anything.
That parameter should make performance worse, which is why that's part of the test-case.
Cheers, Gopal
