[SPARK-55038][SQL] Fix wrong results for array_agg(DISTINCT) with AQE… #54021
+34
−8
Conversation
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
… and sort-based aggregation Sort the output array in CollectList.eval() and CollectSet.eval() to ensure deterministic element ordering. This fixes incorrect results when sort-based aggregation fallback merges partial buffers, producing arrays with the same elements in different orders that then fail equality comparisons in joins used for decorrelation. Both functions document that their output order is non-deterministic, so sorting does not violate their contract.
JIRA Issue Information=== Bug SPARK-55038 === This comment was automatically generated by GitHub Actions |
…t_list Update expected test outputs in CliSuite to reflect the new sorted ordering of collect_list results.
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Outdated suggestions cannot be applied.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.
What changes were proposed in this pull request?
Sort the output array in CollectList.eval() and CollectSet.eval() using
PhysicalDataType.ordering(child.dataType) to ensure deterministic element ordering. This
follows the same pattern used by CollectTopK.eval().
The change affects two methods in collect.scala:
Golden SQL test result files (group-by.sql.out, scalar-subquery-select.sql.out) are
updated to reflect the now-sorted output.
Why are the changes needed?
When AQE is enabled with spark.sql.objectHashAggregate.sortBased.fallbackThreshold=1,
array_agg(DISTINCT) produces wrong results in correlated subqueries. The root cause:
distinct columns to grouping keys in AggUtils.planAggregateWithOneDistinct.
buffers, causing non-deterministic element order across runs.
compare array results for equality.
elements in different orders are treated as unequal, causing the join to produce zero
matches instead of the correct result.
Both collect_list and collect_set explicitly document that their output order is
non-deterministic, so sorting does not violate their contract.
Does this PR introduce any user-facing change?
Yes. collect_list and collect_set now return elements in sorted order instead of
insertion order. Both functions already document that their output order is
non-deterministic and should not be relied upon, so this is a bug fix rather than a
behavior change in terms of the documented API contract.
Before: SELECT collect_list(col) FROM VALUES (1), (2), (1) AS tab(col) → [1,2,1]
After: SELECT collect_list(col) FROM VALUES (1), (2), (1) AS tab(col) → [1,1,2]
How was this patch tested?
consistent equality results with AQE enabled and sortBased.fallbackThreshold=1.
updated expected output.
Was this patch authored or co-authored using generative AI tooling?
No