Hyunsik Choi created TAJO-943:
---------------------------------

             Summary: Preaggregation can be disabled by optimization hint or 
session variable
                 Key: TAJO-943
                 URL: https://issues.apache.org/jira/browse/TAJO-943
             Project: Tajo
          Issue Type: Improvement
          Components: distributed query plan
            Reporter: Hyunsik Choi
            Priority: Minor


Aggregation operators requires two phases and one shuffle as follows:
 * preaggregation -> hash shuffle -> aggregation

Preaggregation usually reduces the intermediate volume between two stages. It 
usually is an effective way to speed up aggregation operation.

But, in some case where the number of distinct grouping keys are lager, 
preaggregation may not reduce the intermediate data while consuming expensive 
operation costs.

It would be useful if we disable preaggregation through some session variable 
or optimization hint.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Reply via email to