Copy limits before repartitions

datafusion/common/src/config.rs

@@ -922,6 +922,11 @@ config_namespace! {
         /// past window functions, if possible
         pub enable_window_limits: bool, default = true
 
+        /// When set to true, the optimizer will push TopK (Sort with fetch)
+        /// below hash repartition when the partition key is a prefix of the
+        /// sort key, reducing data volume before the shuffle.
+        pub enable_topk_repartition: bool, default = true
+
         /// When set to true, the optimizer will attempt to push down TopK dynamic filters
         /// into the file scan phase.
         pub enable_topk_dynamic_filter_pushdown: bool, default = true

datafusion/core/Cargo.toml

@@ -228,6 +228,10 @@ name = "struct_query_sql"
 harness = false
 name = "window_query_sql"
 
+[[bench]]
+harness = false
+name = "topk_repartition"
+
 [[bench]]
 harness = false
 name = "scalar"

datafusion/core/benches/topk_repartition.rs

@@ -0,0 +1,90 @@
+// Licensed to the Apache Software Foundation (ASF) under one
+// or more contributor license agreements.  See the NOTICE file
+// distributed with this work for additional information
+// regarding copyright ownership.  The ASF licenses this file
+// to you under the Apache License, Version 2.0 (the
+// "License"); you may not use this file except in compliance
+// with the License.  You may obtain a copy of the License at
+//
+//   http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing,
+// software distributed under the License is distributed on an
+// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+// KIND, either express or implied.  See the License for the
+// specific language governing permissions and limitations
+// under the License.
+
+//! Benchmark for the TopKRepartition optimizer rule.
+//!
+//! Measures the benefit of pushing TopK (Sort with fetch) below hash
+//! repartition when running partitioned window functions with LIMIT.
+
+mod data_utils;
+
+use criterion::{BenchmarkId, Criterion, criterion_group, criterion_main};
+use data_utils::create_table_provider;
+use datafusion::prelude::{SessionConfig, SessionContext};
+use parking_lot::Mutex;
+use std::hint::black_box;
+use std::sync::Arc;
+use tokio::runtime::Runtime;
+
+#[expect(clippy::needless_pass_by_value)]
+fn query(ctx: Arc<Mutex<SessionContext>>, rt: &Runtime, sql: &str) {
+    let df = rt.block_on(ctx.lock().sql(sql)).unwrap();
+    black_box(rt.block_on(df.collect()).unwrap());
+}
+
+fn create_context(
+    partitions_len: usize,
+    target_partitions: usize,
+    enable_topk_repartition: bool,
+) -> Arc<Mutex<SessionContext>> {
+    let array_len = 1024 * 1024;
+    let batch_size = 8 * 1024;
+    let mut config = SessionConfig::new().with_target_partitions(target_partitions);
+    config.options_mut().optimizer.enable_topk_repartition = enable_topk_repartition;
+    let ctx = SessionContext::new_with_config(config);
+    let rt = Runtime::new().unwrap();
+    rt.block_on(async {
+        let provider =
+            create_table_provider(partitions_len, array_len, batch_size).unwrap();
+        ctx.register_table("t", provider).unwrap();
+    });
+    Arc::new(Mutex::new(ctx))
+}
+
+fn criterion_benchmark(c: &mut Criterion) {
+    let rt = Runtime::new().unwrap();
+
+    let limits = [10, 1_000, 10_000, 100_000];
+    let scans = 16;
+    let target_partitions = 4;
+
+    let group = format!("topk_repartition_{scans}_to_{target_partitions}");
+    let mut group = c.benchmark_group(group);
+    for limit in limits {
+        let sql = format!(
+            "SELECT \
+                SUM(f64) OVER (PARTITION BY u64_narrow ORDER BY u64_wide ROWS UNBOUNDED PRECEDING) \
+                FROM t \
+                ORDER BY u64_narrow, u64_wide \
+                LIMIT {limit}"
+        );
+
+        let ctx_disabled = create_context(scans, target_partitions, false);
+        group.bench_function(BenchmarkId::new("disabled", limit), |b| {
+            b.iter(|| query(ctx_disabled.clone(), &rt, &sql))
+        });
+
+        let ctx_enabled = create_context(scans, target_partitions, true);
+        group.bench_function(BenchmarkId::new("enabled", limit), |b| {
+            b.iter(|| query(ctx_enabled.clone(), &rt, &sql))
+        });
+    }
+    group.finish();
+}
+
+criterion_group!(benches, criterion_benchmark);
+criterion_main!(benches);

datafusion/physical-optimizer/src/lib.rs

@@ -42,6 +42,7 @@ pub use datafusion_pruning as pruning;
 pub mod pushdown_sort;
 pub mod sanity_checker;
 pub mod topk_aggregation;
+pub mod topk_repartition;
 pub mod update_aggr_exprs;
 pub mod utils;
 

datafusion/physical-optimizer/src/optimizer.rs

@@ -33,6 +33,7 @@ use crate::output_requirements::OutputRequirements;
 use crate::projection_pushdown::ProjectionPushdown;
 use crate::sanity_checker::SanityCheckPlan;
 use crate::topk_aggregation::TopKAggregation;
+use crate::topk_repartition::TopKRepartition;
 use crate::update_aggr_exprs::OptimizeAggregateOrder;
 
 use crate::limit_pushdown_past_window::LimitPushPastWindows;
@@ -141,6 +142,11 @@ impl PhysicalOptimizer {
             // replacing operators with fetching variants, or adding limits
             // past operators that support limit pushdown.
             Arc::new(LimitPushdown::new()),
+            // TopKRepartition pushes TopK (Sort with fetch) below Hash
+            // repartition when the partition key is a prefix of the sort key.
+            // This reduces data volume before a hash shuffle. It must run
+            // after LimitPushdown so that the TopK already exists on the SortExec.
+            Arc::new(TopKRepartition::new()),
             // The ProjectionPushdown rule tries to push projections towards
             // the sources in the execution plan. As a result of this process,
             // a projection can disappear if it reaches the source providers, and