arrow-select: optimise coalesced takes for primitive and view arrays#9758
Open
ClSlaid wants to merge 3 commits intoapache:mainfrom
Open
arrow-select: optimise coalesced takes for primitive and view arrays#9758ClSlaid wants to merge 3 commits intoapache:mainfrom
ClSlaid wants to merge 3 commits intoapache:mainfrom
Conversation
## What - add a direct `push_batch_with_indices` path in `BatchCoalescer` for primitive, `Utf8View`, and `BinaryView` columns when the index array is integer typed and non-null - teach the coalescer internals to copy indexed primitive and view values directly into in-progress output buffers instead of materialising an intermediate taken `RecordBatch` - add dedicated `take:` coalesce benchmarks and new indexed coalescing tests for primitive, `Utf8View`, and `BinaryView` inputs ## How - route supported batches through a direct indices path that chunks the input indices across coalesced output batch boundaries and reuses the existing in-progress array builders - specialise `InProgressPrimitiveArray::copy_indices` to gather values and build the taken null mask directly from the source array - specialise `InProgressByteViewArray::copy_indices` to gather selected views and nulls directly, compute buffer compaction from the selected views, and lazily compute whole-array buffer usage only when the row-copy path needs it - keep unsupported types on the existing `take_record_batch` fallback so the optimisation only applies where the benchmark data shows it is profitable ## Why It Works - the previous `push_batch_with_indices` implementation always paid to allocate and populate a temporary taken `RecordBatch` before coalescing - for primitive and view arrays, the coalescer can write the selected rows straight into its output builders, avoiding that extra batch materialisation and the extra copy it implies - the view-array path remains safe because it preserves the existing reuse-vs-compact behaviour, but bases sparse compaction decisions on the actually selected views rather than the whole source batch ## Tests And Validation - added indexed coalescing tests for mixed primitive, mixed `Utf8View`, mixed `BinaryView`, and `Utf8` fallback behaviour in `arrow-select/src/coalesce.rs` - added `take:` coalesce benchmarks in `arrow/benches/coalesce_kernels.rs` covering primitive, `Utf8View`, `BinaryView`, `Utf8`, and dictionary-backed schemas - validated with: - `cargo test -p arrow-select coalesce --lib` - `cargo clippy -p arrow-select --lib --tests -- -D warnings` - `cargo clippy -p arrow --bench coalesce_kernels --features test_utils -- -D warnings` ## Benchmark Summary - `take: primitive, 8192, nulls: 0, selectivity: 0.01`: `3.5194-3.5796 ms` -> `1.8780-1.9136 ms` - `take: primitive, 8192, nulls: 0.1, selectivity: 0.01`: `5.5208-5.5708 ms` -> `4.0016-4.1647 ms` - `take: primitive, 8192, nulls: 0, selectivity: 0.001`: `23.684-23.813 ms` -> `5.9713-6.0137 ms` - `take: single_utf8view, 8192, nulls: 0, selectivity: 0.01`: `3.0301-3.0830 ms` -> `2.4513-2.4854 ms` - `take: mixed_utf8view (max_string_len=20), 8192, nulls: 0, selectivity: 0.01`: `1.8643-1.8823 ms` -> `1.2706-1.2856 ms` - `take: single_binaryview, 8192, nulls: 0, selectivity: 0.01`: `3.1346-3.2991 ms` -> `2.7578-2.8539 ms` - `take: mixed_binaryview (max_string_len=20), 8192, nulls: 0, selectivity: 0.01`: `1.9634-2.0215 ms` -> `1.4117-1.4383 ms` Signed-off-by: cl <cailue@apache.org>
Contributor
Author
Dandandan
reviewed
Apr 18, 2026
Dandandan
reviewed
Apr 18, 2026
Signed-off-by: 蔡略 <cailue@apache.org>
ClSlaid
force-pushed
the
optimize-pr-8991-followup
branch
from
0a6096a to
0bdc485
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
BatchCoalescer::push_batch_with_indicespath for primitive,Utf8View, andBinaryViewcolumns when the indices are integer typed and non-nullRecordBatchtake_record_batchfallback; benchmark work on this branch showed widening the direct path beyond primitive and view arrays regressedUtf8and dictionary-backed casesTesting
cargo test -p arrow-select coalesce --libcargo clippy -p arrow-select --lib --tests -- -D warningscargo clippy -p arrow --bench coalesce_kernels --features test_utils -- -D warningscargo clippy --workspace --all-targets -- -D warningsBenchmarks
take: primitive, 8192, nulls: 0, selectivity: 0.01:3.5194-3.5796 ms->1.8780-1.9136 mstake: primitive, 8192, nulls: 0.1, selectivity: 0.01:5.5208-5.5708 ms->4.0016-4.1647 mstake: primitive, 8192, nulls: 0, selectivity: 0.001:23.684-23.813 ms->5.9713-6.0137 mstake: single_utf8view, 8192, nulls: 0, selectivity: 0.01:3.0301-3.0830 ms->2.4513-2.4854 mstake: mixed_utf8view (max_string_len=20), 8192, nulls: 0, selectivity: 0.01:1.8643-1.8823 ms->1.2706-1.2856 mstake: single_binaryview, 8192, nulls: 0, selectivity: 0.01:3.1346-3.2991 ms->2.7578-2.8539 mstake: mixed_binaryview (max_string_len=20), 8192, nulls: 0, selectivity: 0.01:1.9634-2.0215 ms->1.4117-1.4383 ms