gianm opened a new pull request #9648: SQL: More straightforward handling of join planning. URL: https://github.com/apache/druid/pull/9648 Two major changes that simplify how joins are planned: 1) Stop using JoinProjectTransposeRule as a way of guiding subquery removal. Instead, add logic to DruidJoinRule that identifies removable subqueries and removes them at the point of creating a DruidJoinQueryRel. This approach reduces the size of the planning space and allows the planner to complete quickly. 2) Remove rules that reorder joins. Not because of an impact on the planning time (it seems minimal), but because the decisions that the planner was making in the new tests were sometimes worse than the user-provided order. I think we'll need to go with the user-provided order for now, and revisit reordering when we can add more smarts to the cost estimator. A third change updates numeric ExprEval classes to store their value as a boxed type that corresponds to what it is supposed to be. This is useful because it affects the behavior of "asString", and is included in this patch because it is needed for the new test "testInnerJoinTwoLookupsToTableUsingNumericColumnInReverse". This test relies on CAST('6', 'DOUBLE') stringifying to "6.0" like an actual double would. Fixes #9646.
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
