feat(optimizer): Plan DISTINCT aggregation with nested GroupBy - #1806
Open
kagamiori wants to merge 3 commits into
Open
feat(optimizer): Plan DISTINCT aggregation with nested GroupBy#1806kagamiori wants to merge 3 commits into
kagamiori wants to merge 3 commits into
Conversation
Summary: Extract aggregate input preparation and physical stage selection so DISTINCT lowering can reuse them before aggregate stage selection without recursively replanning inputs. Move the existing pre-grouped-key calculation to shared physical properties without changing its behavior. Differential Revision: D117896114
) Summary: Distributed v2 queries with `MarkDistinct` now choose marker distribution before selecting split vs. single steps for the aggregation. Previously, the late expansion pass inserted `MarkDistinct` after physical planning, leaving the DISTINCT aggregation to always take single-stage. Moves DISTINCT lowering into physical aggregate planning. Plans with one worker and one driver, or with pre-grouped input, continue using native DISTINCT. Removes the separate late expansion pass. Differential Revision: D117940699
Summary: V2 now plans aggregates that share one unfiltered DISTINCT argument set as an inner grouping Aggregate followed by a regular result Aggregate. This reduces rows before the final grouping exchange instead of carrying marker columns through the result aggregation. The inner and outer Aggregates select their physical stages independently. Other DISTINCT shapes continue to use `MarkDistinct`, while single-driver and pre-grouped inputs retain native DISTINCT. This PR also fixes a correctness bug in V1 where a keyless inner Aggregate used to change the empty-input result. Differential Revision: D117954163
meta-cla
Bot
added
the
CLA Signed
|
@kagamiori has exported this pull request. If you are a Meta employee, you can view the originating Diff in D117954163. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary:
V2 now plans aggregates that share one unfiltered DISTINCT argument set as an inner grouping Aggregate followed by a regular result Aggregate. This reduces rows before the final grouping exchange instead of carrying marker columns through the result aggregation.
The inner and outer Aggregates select their physical stages independently. Other DISTINCT shapes continue to use
MarkDistinct, while single-driver and pre-grouped inputs retain native DISTINCT.This PR also fixes a correctness bug in V1 where a keyless inner Aggregate used to change the empty-input result.
Differential Revision: D117954163