fix: Streaming aggregation produces wrong result with grouping sets

[2010YOUY01]

Which issue does this PR close?

• Closes Streaming aggregation produces wrong result with grouping sets #24421

Rationale for this change

The aggregation planning will enable ordering optimization when the grouping sets is available, however they're not compatible now.

I'm also not sure if we want to enable ordering optimization for grouping sets in the future, since this seems requires extra implementation complexity.

This PR disables ordering/streaming aggregation optimization when grouping set is present.

Reference for ordering optimization in aggregation:

datafusion/datafusion/physical-plan/src/aggregates/ordered_partial_stream.rs

Line 62 in 710e74e

/// # Order-based Optimization

What changes are included in this PR?

• 80cb8ee: reproducer that fails without the fix
• 5673f3e: fix. It disables ordering optimization when grouping set is present

Are these changes tested?

slt

Are there any user-facing changes?

no

[codecov-commenter]

Codecov Report

❌ Patch coverage is 96.29630% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 81.23%. Comparing base (1f0615a) to head (5673f3e).
⚠️ Report is 33 commits behind head on main.

Files with missing lines	Patch %	Lines
datafusion/physical-plan/src/aggregates/mod.rs	96.29%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #24422      +/-   ##
==========================================
+ Coverage   81.17%   81.23%   +0.05%     
==========================================
  Files        1109     1112       +3     
  Lines      388117   390652    +2535     
  Branches   388117   390652    +2535     
==========================================
+ Hits       315071   317352    +2281     
- Misses      54504    54660     +156     
- Partials    18542    18640      +98     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[zhuqi-lucas]

Do we need to add some comments or todo for this?

[2010YOUY01]

Updated, thanks!

[alamb]

I'm also not sure if we want to enable ordering optimization for grouping sets in the future, since this seems requires extra implementation complexity.

I don't think so - as there is no one ordering I think that will satisfy all the groupings

[alamb]

Thanks @2010YOUY01 and @zhuqi-lucas

Is this fixing a regression in df 55 (aka there are potentially wrong results in df 55 that were correct in df 54)? If so, I'll add it to the df 55.1 tracking ticket

• #24462

[2010YOUY01]

Thanks @2010YOUY01 and @zhuqi-lucas

Is this fixing a regression in df 55 (aka there are potentially wrong results in df 55 that were correct in df 54)? If so, I'll add it to the df 55.1 tracking ticket

• Release DataFusion 55.1.0 (minor/patch) Release (Sep 2026)  #24462

@alamb No, I've checked this bug can also be reproduced in datafusion 54.0.0, and I believe this bug has always been there.

---

Thank you for the review @zhuqi-lucas and @alamb , let's merge it.