Support planning logical range repartitioning

[gene-bordegaray]

Will Update With More Depth

Part of #22395.

DataFusion now has a logical Partitioning::Range representation, but LogicalPlan::Repartition with range partitioning is rejected during physical planning.

Add planner support to lower logical range repartitioning to an appropriate physical plan once range repartition execution support is available.

[saadtajwar]

Hey @gene-bordegaray ! If this is still open I'm happy to take it and start digging around!

[saadtajwar]

take

[saadtajwar]

@gene-bordegaray Hmm, I definitely could be mistaken here but it looks like in the below that logical range repartitioning is now supported during physical planning:

    async fn map_logical_node_to_physical(...) {
....
                    LogicalPartitioning::Range(range) => {
                        let sort_exprs = create_physical_sort_exprs(
                            range.ordering(),
                            input_dfschema,
                            execution_props,
                        )?;
                        let ordering = LexOrdering::new(sort_exprs).ok_or_else(|| {
                            internal_datafusion_err!(
                                "Range repartitioning requires non-empty ordering"
                            )
                        })?;
                        Partitioning::Range(RangePartitioning::try_new(
                            ordering,
                            range.split_points().to_vec(),
                        )?)
                    }
...
}

With the test physical_planner::tests::logical_range_repartition_plans_output_partitioning passing as well! Is the above what this issue was encompassing, or is there something else that I might be missing here? Thanks again for the help! 😁

[gene-bordegaray]

yes forgot to close this, but the actual physical implementation for repartitioning for Range is not implemented if you would like to look at that @saadtajwar 👍

[gene-bordegaray]

completed in #22777

[saadtajwar]

yes forgot to close this, but the actual physical implementation for repartitioning for Range is not implemented if you would like to look at that @saadtajwar 👍

Sounds good, yeah for sure! I'll make an issue for that if there isn't one already