Skip to content

perf: Use DataFusion's count_udaf instead of SUM(IF(expr IS NOT NULL, 1, 0)) #2407

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
andygrove merged 7 commits into from
Sep 16, 2025
Merged

perf: Use DataFusion's count_udaf instead of SUM(IF(expr IS NOT NULL, 1, 0)) #2407

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
025d4bc
refactor fuzz test
andygrove Sep 16, 2025
6f0a12b
link to issue
andygrove Sep 16, 2025
b2eec39
Use DF count_udaf
andygrove Sep 16, 2025
714e595
add test to CI
andygrove Sep 16, 2025
9adad89
Merge remote-tracking branch 'apache/main' into use-df-count
andygrove Sep 16, 2025
ec0c9e2
update benchmark
andygrove Sep 16, 2025
a25f0a8
update benchmark
andygrove Sep 16, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions .gitignore
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,3 +17,4 @@ dev/dist
apache-rat-*.jar
venv
dev/release/comet-rm/workdir
spark/benchmarks
25 changes: 2 additions & 23 deletions native/core/src/execution/planner.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,6 +30,7 @@ use crate::{
use arrow::compute::CastOptions;
use arrow::datatypes::{DataType, Field, Schema, TimeUnit, DECIMAL128_MAX_PRECISION};
use datafusion::functions_aggregate::bit_and_or_xor::{bit_and_udaf, bit_or_udaf, bit_xor_udaf};
use datafusion::functions_aggregate::count::count_udaf;
use datafusion::functions_aggregate::min_max::max_udaf;
use datafusion::functions_aggregate::min_max::min_udaf;
use datafusion::functions_aggregate::sum::sum_udaf;
Expand Down Expand Up @@ -1904,35 +1905,13 @@ impl PhysicalPlanner {
match spark_expr.expr_struct.as_ref().unwrap() {
AggExprStruct::Count(expr) => {
assert!(!expr.children.is_empty());
// Using `count_udaf` from Comet is exceptionally slow for some reason, so
// as a workaround we translate it to `SUM(IF(expr IS NOT NULL, 1, 0))`
// https://github.com/apache/datafusion-comet/issues/744

let children = expr
.children
.iter()
.map(|child| self.create_expr(child, Arc::clone(&schema)))
.collect::<Result<Vec<_>, _>>()?;

// create `IS NOT NULL expr` and join them with `AND` if there are multiple
let not_null_expr: Arc<dyn PhysicalExpr> = children.iter().skip(1).fold(
Arc::new(IsNotNullExpr::new(Arc::clone(&children[0]))) as Arc<dyn PhysicalExpr>,
|acc, child| {
Arc::new(BinaryExpr::new(
acc,
DataFusionOperator::And,
Arc::new(IsNotNullExpr::new(Arc::clone(child))),
))
},
);

let child = Arc::new(IfExpr::new(
not_null_expr,
Arc::new(Literal::new(ScalarValue::Int64(Some(1)))),
Arc::new(Literal::new(ScalarValue::Int64(Some(0)))),
));

AggregateExprBuilder::new(sum_udaf(), vec![child])
AggregateExprBuilder::new(count_udaf(), children)
.schema(schema)
.alias("count")
.with_ignore_nulls(false)
Expand Down
34 changes: 4 additions & 30 deletions spark/src/test/scala/org/apache/spark/sql/benchmark/CometAggregateBenchmark.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -64,13 +64,7 @@ object CometAggregateBenchmark extends CometBenchmarkBase {
spark.sql(query).noop()
}

benchmark.addCase(s"SQL Parquet - Comet (Scan) ($aggregateFunction)") { _ =>
withSQLConf(CometConf.COMET_ENABLED.key -> "true") {
spark.sql(query).noop()
}
}
Comment on lines -67 to -71
Copy link
Member Author

@andygrove andygrove Sep 16, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It no longer makes sense to benchmark Comet with Scan vs Scan & Exec configurations, especially for aggregates. Comet scans won't accelerate aggregates.


benchmark.addCase(s"SQL Parquet - Comet (Scan, Exec) ($aggregateFunction)") { _ =>
benchmark.addCase(s"SQL Parquet - Comet ($aggregateFunction)") { _ =>
withSQLConf(
CometConf.COMET_ENABLED.key -> "true",
CometConf.COMET_EXEC_ENABLED.key -> "true") {
Expand Down Expand Up @@ -111,13 +105,7 @@ object CometAggregateBenchmark extends CometBenchmarkBase {
spark.sql(query).noop()
}

benchmark.addCase(s"SQL Parquet - Comet (Scan) ($aggregateFunction)") { _ =>
withSQLConf(CometConf.COMET_ENABLED.key -> "true") {
spark.sql(query).noop()
}
}

benchmark.addCase(s"SQL Parquet - Comet (Scan, Exec) ($aggregateFunction)") { _ =>
benchmark.addCase(s"SQL Parquet - Comet ($aggregateFunction)") { _ =>
withSQLConf(
CometConf.COMET_ENABLED.key -> "true",
CometConf.COMET_EXEC_ENABLED.key -> "true") {
Expand Down Expand Up @@ -153,15 +141,7 @@ object CometAggregateBenchmark extends CometBenchmarkBase {
spark.sql(query).noop()
}

benchmark.addCase(s"SQL Parquet - Comet (Scan) ($aggregateFunction)") { _ =>
withSQLConf(
CometConf.COMET_ENABLED.key -> "true",
CometConf.COMET_MEMORY_OVERHEAD.key -> "1G") {
spark.sql(query).noop()
}
}

benchmark.addCase(s"SQL Parquet - Comet (Scan, Exec) ($aggregateFunction)") { _ =>
benchmark.addCase(s"SQL Parquet - Comet ($aggregateFunction)") { _ =>
withSQLConf(
CometConf.COMET_ENABLED.key -> "true",
CometConf.COMET_EXEC_ENABLED.key -> "true",
Expand Down Expand Up @@ -198,13 +178,7 @@ object CometAggregateBenchmark extends CometBenchmarkBase {
spark.sql(query).noop()
}

benchmark.addCase(s"SQL Parquet - Comet (Scan) ($aggregateFunction)") { _ =>
withSQLConf(CometConf.COMET_ENABLED.key -> "true") {
spark.sql(query).noop()
}
}

benchmark.addCase(s"SQL Parquet - Comet (Scan, Exec) ($aggregateFunction)") { _ =>
benchmark.addCase(s"SQL Parquet - Comet ($aggregateFunction)") { _ =>
withSQLConf(
CometConf.COMET_ENABLED.key -> "true",
CometConf.COMET_EXEC_ENABLED.key -> "true") {
Expand Down
Loading