Hi Marieke, Usually for a fledged SQL engine built on top of Calcite, it uses both planner. And we have Program[1] abstraction to organize multi stages to plan a query. You can take a look at Calcite's default implementation here[2].
[1] https://github.com/apache/calcite/blob/master/core/src/main/java/org/apache/calcite/tools/Program.java [2] https://github.com/apache/calcite/blob/master/core/src/main/java/org/apache/calcite/prepare/Prepare.java#L176 Marieke Gueye <[email protected]> 于2022年4月23日周六 04:50写道： > Hey everybody, > > I created an adapter with some rules pushing down everything (Projects, > Filter,..) to a single Rel, let's call it CustomRel: > > So for example my Project Rule basically transforms: > LogicalProject(R_NAME=[$0]) > CustomRel(projects=[[$0, $1]]) > to > CustomRel(projects=[[$0]]) > > I also have a CustomRelToEnumerableConverter & > CustomRelToEnumerableConverterRule that converts the trait set from my > custom convention (in the CustomRel) to an Enumerable. > > I am using Avatica with my adapter and realised that the Volcano planner is > not working as I would expect it to. Instead of first pushing down > everything to CustomRel then converting to enumerable, it is pushing down > only one rel above it, applying the convert and then doing some > EnumerableCalc.. Changing the selfCost of CustomRel did not help. > > I also tried overriding the planner (overriding createPlanner from > CalcitePrepareImpl) so it uses a HepPlanner but it looks like the planner > loses the trait information and the converter is never fired. > > Any idea what would be the solution in this case? > > Thanks for your help > -- Best, Benchao Li
