You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@datafusion.apache.org by gi...@apache.org on 2025/12/04 17:34:39 UTC

(datafusion) branch main updated: Update aggregate probe to be locked only if skipping aggregation (#18766)

This is an automated email from the ASF dual-hosted git repository.

github-bot pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/datafusion.git


The following commit(s) were added to refs/heads/main by this push:
     new 388db0e52f Update aggregate probe to be locked only if skipping aggregation (#18766)
388db0e52f is described below

commit 388db0e52f8e31256ba6185cfb9605c1e678c74c
Author: Haresh Khanna <kh...@gmail.com>
AuthorDate: Thu Dec 4 17:34:12 2025 +0000

    Update aggregate probe to be locked only if skipping aggregation (#18766)
    
    ## Which issue does this PR close?
    
    <!--
    We generally require a GitHub issue to be filed for all bug fixes and
    enhancements and this helps us generate change logs for our releases.
    You can link an issue to this PR using the GitHub syntax. For example
    `Closes #123` indicates that this PR will close issue #123.
    -->
    
    - Closes #.
    
    ## Rationale for this change
    
    <!--
    Why are you proposing this change? If this is already explained clearly
    in the issue then this section is not needed.
    Explaining clearly why changes are proposed helps reviewers understand
    your changes and offer better suggestions for fixes.
    -->
    
    If we always set the probe to locked, we let go of the ability to skip
    partial aggregation during processing a future record batch when the
    ratio does meet the required threshold to skip.
    
    ## What changes are included in this PR?
    
    <!--
    There is no need to duplicate the description in the issue here but it
    is sometimes worth providing a summary of the individual changes in this
    PR.
    -->
    
    Lock the aggregation probe only if we are skipping not when just the
    number of rows threshold is met.
    
    ## Are these changes tested?
    
    <!--
    We typically require tests for all PRs in order to:
    1. Prevent the code from being accidentally broken by subsequent changes
    2. Serve as another way to document the expected behavior of the code
    
    If tests are not included in your PR, please explain why (for example,
    are they covered by existing tests)?
    -->
    
    This change should be covered by existing aggregate tests.
    
    ## Are there any user-facing changes?
    
    <!--
    If there are user-facing changes then we may require documentation to be
    updated before approving the PR.
    -->
    
    <!--
    If there are any breaking changes to public APIs, please add the `api
    change` label.
    -->
    No
---
 .../physical-plan/src/aggregates/row_hash.rs       | 151 ++++++++++++++++++++-
 1 file changed, 150 insertions(+), 1 deletion(-)

diff --git a/datafusion/physical-plan/src/aggregates/row_hash.rs b/datafusion/physical-plan/src/aggregates/row_hash.rs
index 3a1560d9ef..615ed28174 100644
--- a/datafusion/physical-plan/src/aggregates/row_hash.rs
+++ b/datafusion/physical-plan/src/aggregates/row_hash.rs
@@ -194,7 +194,9 @@ impl SkipAggregationProbe {
         if self.input_rows >= self.probe_rows_threshold {
             self.should_skip = self.num_groups as f64 / self.input_rows as f64
                 >= self.probe_ratio_threshold;
-            self.is_locked = true;
+            // Set is_locked to true only if we have decided to skip, otherwise we can try to skip
+            // during processing the next record_batch.
+            self.is_locked = self.should_skip;
         }
     }
 
@@ -1280,6 +1282,7 @@ impl GroupedHashAggregateStream {
 #[cfg(test)]
 mod tests {
     use super::*;
+    use crate::execution_plan::ExecutionPlan;
     use crate::test::TestMemoryExec;
     use arrow::array::{Int32Array, Int64Array};
     use arrow::datatypes::{DataType, Field, Schema};
@@ -1395,4 +1398,150 @@ mod tests {
 
         Ok(())
     }
+
+    #[tokio::test]
+    async fn test_skip_aggregation_probe_not_locked_until_skip() -> Result<()> {
+        // Test that the probe is not locked until we actually decide to skip.
+        // This allows us to continue evaluating the skip condition across multiple batches.
+        //
+        // Scenario:
+        // - Batch 1: Hits rows threshold but NOT ratio threshold (low cardinality) -> don't skip
+        // - Batch 2: Now hits ratio threshold (high cardinality) -> skip
+        //
+        // Without the fix, the probe would be locked after batch 1, preventing the skip
+        // decision from being made on batch 2.
+
+        let schema = Arc::new(Schema::new(vec![
+            Field::new("group_col", DataType::Int32, false),
+            Field::new("value_col", DataType::Int32, false),
+        ]));
+
+        // Configure thresholds:
+        // - probe_rows_threshold: 100 rows
+        // - probe_ratio_threshold: 0.8 (80%)
+        let probe_rows_threshold = 100;
+        let probe_ratio_threshold = 0.8;
+
+        // Batch 1: 100 rows with only 10 unique groups
+        // Ratio: 10/100 = 0.1 (10%) < 0.8 -> should NOT skip
+        // This will hit the rows threshold but not the ratio threshold
+        let batch1_rows = 100;
+        let batch1_groups = 10;
+        let mut group_ids_batch1 = Vec::new();
+        for i in 0..batch1_rows {
+            group_ids_batch1.push((i % batch1_groups) as i32);
+        }
+        let values_batch1: Vec<i32> = vec![1; batch1_rows];
+
+        let batch1 = RecordBatch::try_new(
+            Arc::clone(&schema),
+            vec![
+                Arc::new(Int32Array::from(group_ids_batch1)),
+                Arc::new(Int32Array::from(values_batch1)),
+            ],
+        )?;
+
+        // Batch 2: 350 rows with 350 unique NEW groups (starting from group 10)
+        // After batch 2, total: 450 rows, 360 groups
+        // Ratio: 360/450 = 0.8 (80%) >= 0.8 -> SHOULD decide to skip
+        let batch2_rows = 350;
+        let batch2_groups = 350;
+        let group_ids_batch2: Vec<i32> = (batch1_groups..(batch1_groups + batch2_groups))
+            .map(|x| x as i32)
+            .collect();
+        let values_batch2: Vec<i32> = vec![1; batch2_rows];
+
+        let batch2 = RecordBatch::try_new(
+            Arc::clone(&schema),
+            vec![
+                Arc::new(Int32Array::from(group_ids_batch2)),
+                Arc::new(Int32Array::from(values_batch2)),
+            ],
+        )?;
+
+        // Batch 3: This batch should be skipped since we decided to skip after batch 2
+        // 100 rows with 100 unique groups (continuing from where batch 2 left off)
+        let batch3_rows = 100;
+        let batch3_groups = 100;
+        let batch3_start_group = batch1_groups + batch2_groups;
+        let group_ids_batch3: Vec<i32> = (batch3_start_group
+            ..(batch3_start_group + batch3_groups))
+            .map(|x| x as i32)
+            .collect();
+        let values_batch3: Vec<i32> = vec![1; batch3_rows];
+
+        let batch3 = RecordBatch::try_new(
+            Arc::clone(&schema),
+            vec![
+                Arc::new(Int32Array::from(group_ids_batch3)),
+                Arc::new(Int32Array::from(values_batch3)),
+            ],
+        )?;
+
+        let input_partitions = vec![vec![batch1, batch2, batch3]];
+
+        let runtime = RuntimeEnvBuilder::default().build_arc()?;
+        let mut task_ctx = TaskContext::default().with_runtime(runtime);
+
+        // Configure skip aggregation settings
+        let mut session_config = task_ctx.session_config().clone();
+        session_config = session_config.set(
+            "datafusion.execution.skip_partial_aggregation_probe_rows_threshold",
+            &datafusion_common::ScalarValue::UInt64(Some(probe_rows_threshold)),
+        );
+        session_config = session_config.set(
+            "datafusion.execution.skip_partial_aggregation_probe_ratio_threshold",
+            &datafusion_common::ScalarValue::Float64(Some(probe_ratio_threshold)),
+        );
+        task_ctx = task_ctx.with_session_config(session_config);
+        let task_ctx = Arc::new(task_ctx);
+
+        // Create aggregate: COUNT(*) GROUP BY group_col
+        let group_expr = vec![(col("group_col", &schema)?, "group_col".to_string())];
+        let aggr_expr = vec![Arc::new(
+            AggregateExprBuilder::new(count_udaf(), vec![col("value_col", &schema)?])
+                .schema(Arc::clone(&schema))
+                .alias("count_value")
+                .build()?,
+        )];
+
+        let exec = TestMemoryExec::try_new(&input_partitions, Arc::clone(&schema), None)?;
+        let exec = Arc::new(TestMemoryExec::update_cache(&Arc::new(exec)));
+
+        // Use Partial mode
+        let aggregate_exec = AggregateExec::try_new(
+            AggregateMode::Partial,
+            PhysicalGroupBy::new_single(group_expr),
+            aggr_expr,
+            vec![None],
+            exec,
+            Arc::clone(&schema),
+        )?;
+
+        // Execute and collect results
+        let mut stream =
+            GroupedHashAggregateStream::new(&aggregate_exec, &Arc::clone(&task_ctx), 0)?;
+        let mut results = Vec::new();
+
+        while let Some(result) = stream.next().await {
+            let batch = result?;
+            results.push(batch);
+        }
+
+        // Check that skip aggregation actually happened
+        // The key metric is skipped_aggregation_rows
+        let metrics = aggregate_exec.metrics().unwrap();
+        let skipped_rows = metrics
+            .sum_by_name("skipped_aggregation_rows")
+            .map(|m| m.as_usize())
+            .unwrap_or(0);
+
+        // We expect batch 3's rows to be skipped (100 rows)
+        assert_eq!(
+            skipped_rows, batch3_rows,
+            "Expected batch 3's rows ({batch3_rows}) to be skipped",
+        );
+
+        Ok(())
+    }
 }


---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@datafusion.apache.org
For additional commands, e-mail: commits-help@datafusion.apache.org