I’m trying to benchmark mapd-core data transferring and query execution and one of thing that I want to tune is execution option ‘output_columnar_hint’. I changed it to enable state but it seems nothing changed for most queries I tried. But some queries (for example, picked from test case, “SELECT dd AS key1, COUNT(*) AS value1 FROM test GROUP BY key1 HAVING key1 IS NOT NULL ORDER BY key1, value1 DESC LIMIT 12”) aren’t work (F1108 06:26:04.447276 12206 GroupByAndAggregate.cpp:2879] Check failed: !outputColumnar()).
Second question is about ResultSet data representation. By default mapd-core places query result data into row-oriented data arrays, but it intends convert it to columnar representation between subqueries because query expects columnar data as input. Why you are using row-oriented output rather than columnar? Probably it will be changed in future?