alpaylan opened a new issue, #26626:
URL: https://github.com/apache/datafusion/issues/26626
### Describe the bug
With `datafusion.execution.target_partitions = 1`, a query with two window
functions over an input with more than one partition fails to plan with a
`SanityCheckPlan` distribution error. The same query runs with
`target_partitions = 4`, and it runs with only one of the two window functions.
`target_partitions` is a parallelism setting, so it should not decide
whether a valid query plans.
### To Reproduce
`datafusion-cli`:
```sql
SET datafusion.execution.target_partitions = 1;
SELECT count(*) OVER (PARTITION BY a + NULL) AS x,
count(*) OVER (PARTITION BY a) AS y
FROM (SELECT 1 AS a UNION ALL SELECT 2);
```
```
SanityCheckPlan
caused by
Error during planning: Plan: ["WindowAggExec: wdw=[count(Int64(1)) PARTITION
BY [a + NULL] ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING: ...]",
" UnionExec", " ProjectionExec: expr=[1 as a]", " PlaceholderRowExec",
" ProjectionExec: expr=[2 as a]", " PlaceholderRowExec"] does not
satisfy distribution requirements: KeyPartitioned[[NULL]]). Child-0 output
partitioning: UnknownPartitioning(2)
```
### Expected behavior
Two rows, as with `target_partitions = 4`:
```
x,y
2,1
2,1
```
### Additional context
- `main` at c0ec5127ac99 (2026-10-22): fails as above. Release 55.2.0 fails
with the same error.
- Only `target_partitions = 1` is needed; the `UNION ALL` gives the input
two partitions.
- The physical planner builds `WindowAggExec` with `can_repartition = false`
when `target_partitions == 1`, so the window requires `SinglePartition` and
gets a `CoalescePartitionsExec`. `WindowAggExec::replace_children` with
`ChildrenPropertiesMode::Recompute` (used by `with_new_children`,
`datafusion/physical-plan/src/windows/window_agg_exec.rs`) rebuilds the node
with `can_repartition = true` instead of `self.can_repartition`. After sort
enforcement replaces its children, the window requires `KeyPartitioned`, the
coalesce is removed, and the sanity check fails. `BoundedWindowAggExec` keeps
`self.can_repartition` at the same place.
- Related, not the same: #21826 (`InterleaveExec::with_new_children` with
changed child partitioning).
Found with an automated differential/metamorphic testing tool; reproduced on
current `main`.
### Willingness to contribute
I would like to contribute a fix for this bug, but request guidance from the
DataFusion community
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
