gortiz opened a new pull request, #19023:
URL: https://github.com/apache/pinot/pull/19023
## Problem
`PinotAggregateExchangeNodeInsertRule.createPlanWithExchangeDirectAggregation`
(the `is_skip_leaf_stage_group_by` / direct-aggregation path) casts the result
of `generateProjectUnderAggregate` directly to `Aggregate`:
```java
aggRel = (Aggregate) generateProjectUnderAggregate(call, aggRel);
```
But `generateProjectUnderAggregate` rebuilds the aggregate with
`RelBuilder`, which **deduplicates identical aggregate calls** and, when it
does, adds a top `Project` to restore the original output layout (a column
referenced more than once is projected twice). In that case the helper returns
a `Project`-over-`Aggregate`, and the cast throws:
```
java.lang.ClassCastException: class
org.apache.calcite.rel.logical.LogicalProject
cannot be cast to class org.apache.calcite.rel.core.Aggregate
```
This is reachable whenever an aggregate carrying
`is_skip_leaf_stage_group_by` ends up with duplicate aggregate calls — e.g.
after an `AVG` reduction reuses an existing `SUM`, leaving two identical `SUM`
calls that `RelBuilder` then deduplicates.
## Fix
Detect the layout-restoring top `Project`, extract the `Aggregate` beneath
it for the direct-aggregation rewrite, and **re-apply the `Project`** (same
projects and row type) over the resulting direct aggregate so the output
columns are preserved. `createPlan` / `createPlanWithExchangeDirectAggregation`
now return `RelNode`.
## Testing
New `PinotAggregateExchangeNodeInsertRuleTest`:
- `skipLeafWithDuplicateAggregateCallsPreservesLayoutProject` — the
duplicate-agg-call case that previously threw; asserts the plan is a
layout-restoring `Project` over a `PinotLogicalAggregate`.
- `skipLeafWithoutDuplicatesProducesDirectAggregate` — the common single-agg
case still produces a bare direct aggregate.
Existing `QueryCompilationTest` (213 tests) passes unchanged.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
--
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]
