Coalesce batches inside hash join, reuse indices buffer (#18972)

## 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 #18781

## Rationale for this change

Improves the plan / plan readability for queries with joins.

Seems to improve perf as well a bit for more challenging joins: (TPC-H
SF=10, in-memory)

```
│ QQuery 21    │              7948.54 ms │    6340.74 ms │ +1.25x faster │
│ QQuery 22    │                58.30 ms │      49.86 ms │ +1.17x faster │
```

<!--
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.
-->

## 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.
-->

## 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)?
-->

## 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.
-->

---------

Co-authored-by: Marko Milenković <milenkovicm@users.noreply.github.com>

Author: Dandandan

datafusion/core/tests/dataframe/mod.rs

@@ -974,6 +974,10 @@ impl JoinFuzzTestCase {
             if join_tests.contains(&NljHj) {
                 let err_msg_rowcnt = format!("NestedLoopJoinExec and HashJoinExec produced different row counts, batch_size: {batch_size}");
                 assert_eq!(nlj_rows, hj_rows, "{}", err_msg_rowcnt.as_str());
+                if nlj_rows == 0 && hj_rows == 0 {
+                    // both joins returned no rows, skip content comparison
+                    continue;
+                }
 
                 let err_msg_contents = format!("NestedLoopJoinExec and HashJoinExec produced different results, batch_size: {batch_size}");
                 // row level compare if any of joins returns the result

datafusion/core/tests/fuzz_cases/join_fuzz.rs

@@ -812,14 +812,13 @@ async fn test_physical_plan_display_indent_multi_children() {
     assert_snapshot!(
         actual,
         @r"
-    CoalesceBatchesExec: target_batch_size=4096
-      HashJoinExec: mode=Partitioned, join_type=Inner, on=[(c1@0, c2@0)], projection=[c1@0]
-        CoalesceBatchesExec: target_batch_size=4096
-          RepartitionExec: partitioning=Hash([c1@0], 9000), input_partitions=1
-            DataSourceExec: file_groups={1 group: [[ARROW_TEST_DATA/csv/aggregate_test_100.csv]]}, projection=[c1], file_type=csv, has_header=true
-        CoalesceBatchesExec: target_batch_size=4096
-          RepartitionExec: partitioning=Hash([c2@0], 9000), input_partitions=1
-            DataSourceExec: file_groups={1 group: [[ARROW_TEST_DATA/csv/aggregate_test_100.csv]]}, projection=[c1@0 as c2], file_type=csv, has_header=true
+    HashJoinExec: mode=Partitioned, join_type=Inner, on=[(c1@0, c2@0)], projection=[c1@0]
+      CoalesceBatchesExec: target_batch_size=4096
+        RepartitionExec: partitioning=Hash([c1@0], 9000), input_partitions=1
+          DataSourceExec: file_groups={1 group: [[ARROW_TEST_DATA/csv/aggregate_test_100.csv]]}, projection=[c1], file_type=csv, has_header=true
+      CoalesceBatchesExec: target_batch_size=4096
+        RepartitionExec: partitioning=Hash([c2@0], 9000), input_partitions=1
+          DataSourceExec: file_groups={1 group: [[ARROW_TEST_DATA/csv/aggregate_test_100.csv]]}, projection=[c1@0 as c2], file_type=csv, has_header=true
     "
     );
 }

datafusion/core/tests/sql/explain_analyze.rs

@@ -27,7 +27,7 @@ use datafusion_common::error::Result;
 use datafusion_physical_expr::Partitioning;
 use datafusion_physical_plan::{
     async_func::AsyncFuncExec, coalesce_batches::CoalesceBatchesExec,
-    joins::HashJoinExec, repartition::RepartitionExec, ExecutionPlan,
+    repartition::RepartitionExec, ExecutionPlan,
 };
 
 use datafusion_common::tree_node::{Transformed, TransformedResult, TreeNode};
@@ -57,17 +57,16 @@ impl PhysicalOptimizerRule for CoalesceBatches {
         let target_batch_size = config.execution.batch_size;
         plan.transform_up(|plan| {
             let plan_any = plan.as_any();
-            let wrap_in_coalesce = plan_any.downcast_ref::<HashJoinExec>().is_some()
+            let wrap_in_coalesce = plan_any
                 // Don't need to add CoalesceBatchesExec after a round robin RepartitionExec
-                || plan_any
-                    .downcast_ref::<RepartitionExec>()
-                    .map(|repart_exec| {
-                        !matches!(
-                            repart_exec.partitioning().clone(),
-                            Partitioning::RoundRobinBatch(_)
-                        )
-                    })
-                    .unwrap_or(false);
+                .downcast_ref::<RepartitionExec>()
+                .map(|repart_exec| {
+                    !matches!(
+                        repart_exec.partitioning().clone(),
+                        Partitioning::RoundRobinBatch(_)
+                    )
+                })
+                .unwrap_or(false);
 
             if wrap_in_coalesce {
                 Ok(Transformed::yes(Arc::new(CoalesceBatchesExec::new(

datafusion/physical-optimizer/src/coalesce_batches.rs

@@ -1936,7 +1936,12 @@ mod tests {
             div_ceil(9, batch_size)
         };
 
-        assert_eq!(batches.len(), expected_batch_count);
+        // With batch coalescing, we may have fewer batches than expected
+        assert!(
+            batches.len() <= expected_batch_count,
+            "expected at most {expected_batch_count} batches, got {}",
+            batches.len()
+        );
 
         // Inner join output is expected to preserve both inputs order
         allow_duplicates! {
@@ -2016,7 +2021,12 @@ mod tests {
             div_ceil(9, batch_size)
         };
 
-        assert_eq!(batches.len(), expected_batch_count);
+        // With batch coalescing, we may have fewer batches than expected
+        assert!(
+            batches.len() <= expected_batch_count,
+            "expected at most {expected_batch_count} batches, got {}",
+            batches.len()
+        );
 
         // Inner join output is expected to preserve both inputs order
         allow_duplicates! {
@@ -2152,7 +2162,12 @@ mod tests {
             // and filtered later.
             div_ceil(6, batch_size)
         };
-        assert_eq!(batches.len(), expected_batch_count);
+        // With batch coalescing, we may have fewer batches than expected
+        assert!(
+            batches.len() <= expected_batch_count,
+            "expected at most {expected_batch_count} batches, got {}",
+            batches.len()
+        );
 
         // Inner join output is expected to preserve both inputs order
         allow_duplicates! {
@@ -2177,7 +2192,12 @@ mod tests {
             // and filtered later.
             div_ceil(3, batch_size)
         };
-        assert_eq!(batches.len(), expected_batch_count);
+        // With batch coalescing, we may have fewer batches than expected
+        assert!(
+            batches.len() <= expected_batch_count,
+            "expected at most {expected_batch_count} batches, got {}",
+            batches.len()
+        );
 
         // Inner join output is expected to preserve both inputs order
         allow_duplicates! {
@@ -3467,6 +3487,8 @@ mod tests {
         let mut hashes_buffer = vec![0; right.num_rows()];
         create_hashes([&right_keys_values], &random_state, &mut hashes_buffer)?;
 
+        let mut probe_indices_buffer = Vec::new();
+        let mut build_indices_buffer = Vec::new();
         let (l, r, _) = lookup_join_hashmap(
             &join_hash_map,
             &[left_keys_values],
@@ -3475,6 +3497,8 @@ mod tests {
             &hashes_buffer,
             8192,
             (0, None),
+            &mut probe_indices_buffer,
+            &mut build_indices_buffer,
         )?;
 
         let left_ids: UInt64Array = vec![0, 1].into();
@@ -3524,6 +3548,8 @@ mod tests {
         let mut hashes_buffer = vec![0; right.num_rows()];
         create_hashes([&right_keys_values], &random_state, &mut hashes_buffer)?;
 
+        let mut probe_indices_buffer = Vec::new();
+        let mut build_indices_buffer = Vec::new();
         let (l, r, _) = lookup_join_hashmap(
             &join_hash_map,
             &[left_keys_values],
@@ -3532,6 +3558,8 @@ mod tests {
             &hashes_buffer,
             8192,
             (0, None),
+            &mut probe_indices_buffer,
+            &mut build_indices_buffer,
         )?;
 
         // We still expect to match rows 0 and 1 on both sides
@@ -4197,10 +4225,11 @@ mod tests {
                     }
                     _ => div_ceil(expected_resultset_records, batch_size) + 1,
                 };
-                assert_eq!(
-                    batches.len(),
-                    expected_batch_count,
-                    "expected {expected_batch_count} output batches for {join_type} join with batch_size = {batch_size}"
+                // With batch coalescing, we may have fewer batches than expected
+                assert!(
+                    batches.len() <= expected_batch_count,
+                    "expected at most {expected_batch_count} output batches for {join_type} join with batch_size = {batch_size}, got {}",
+                    batches.len()
                 );
 
                 let expected = match join_type {
@@ -4210,7 +4239,17 @@ mod tests {
                     JoinType::LeftAnti => left_empty.to_vec(),
                     _ => common_result.to_vec(),
                 };
-                assert_batches_eq!(expected, &batches);
+                // For anti joins with empty results, we may get zero batches
+                // (with coalescing) instead of one empty batch with schema
+                if batches.is_empty() {
+                    // Verify this is an expected empty result case
+                    assert!(
+                        matches!(join_type, JoinType::RightAnti | JoinType::LeftAnti),
+                        "Unexpected empty result for {join_type} join"
+                    );
+                } else {
+                    assert_batches_eq!(expected, &batches);
+                }
             }
         }
     }
@@ -4473,9 +4512,15 @@ mod tests {
 
         assert_join_metrics!(metrics, 0);
 
-        let expected_null_neq =
-            ["+----+----+", "| n1 | n2 |", "+----+----+", "+----+----+"];
-        assert_batches_eq!(expected_null_neq, &batches_null_neq);
+        // With batch coalescing, empty results may not emit any batches
+        // Check that either we have no batches, or an empty batch with proper schema
+        if batches_null_neq.is_empty() {
+            // This is fine - no output rows
+        } else {
+            let expected_null_neq =
+                ["+----+----+", "| n1 | n2 |", "+----+----+", "+----+----+"];
+            assert_batches_eq!(expected_null_neq, &batches_null_neq);
+        }
 
         Ok(())
     }