[arrow-select] Replace ArrayData with direct Array construction in filter kernels#9986
Open
liamzwbao wants to merge 3 commits into
Open
[arrow-select] Replace ArrayData with direct Array construction in filter kernels#9986liamzwbao wants to merge 3 commits into
ArrayData with direct Array construction in filter kernels#9986liamzwbao wants to merge 3 commits into
Conversation
liamzwbao
changed the title
ArrayData with direct Array constructionArrayData with direct Array construction in filter
liamzwbao
changed the title
ArrayData with direct Array construction in filterArrayData with direct Array construction in filter kernels
Contributor
|
run benchmark filter_kernels |
|
🤖 Arrow criterion benchmark running (GKE) | trigger CPU Details (lscpu)Comparing issue-9298-repalce-array-data-arrow-select (34b1837) to accb1cf (merge-base) diff File an issue against this benchmark runner |
alamb
approved these changes
May 20, 2026
Contributor
alamb
left a comment
alamb
left a comment
There was a problem hiding this comment.
Thank you @liamzwbao -- this is great. It is hard not to love a PR that:
- Makes the code simpler (fewer lines)
- Removes uses of
unsafe - Make things faster
I left a few stylistic comments and am running the benchmarks. Assuming the benchmark results look good I think this PR is good to merge
Thank you again
| let (null_count, nulls) = filter_null_mask(nulls, predicate)?; | ||
| let buffer = BooleanBuffer::new(nulls, 0, predicate.count); | ||
|
|
||
| Some(unsafe { NullBuffer::new_unchecked(buffer, null_count) }) |
Contributor
There was a problem hiding this comment.
can we please add a safety comment here explaining why this is safe:
Suggested change
| Some(unsafe { NullBuffer::new_unchecked(buffer, null_count) }) | |
| // Safety: null_count return from filter_null_mas is correct | |
| Some(unsafe { NullBuffer::new_unchecked(buffer, null_count) }) |
It might also be nice to add a debug assert here to verify in debug builds
debug_assert_eq!(null_count, nulls.num_zeros())| .len(predicate.count) | ||
| .add_buffer(filter.dst_offsets.into()) | ||
| .add_buffer(filter.dst_values.into()); | ||
| let offsets = unsafe { OffsetBuffer::new_unchecked(filter.dst_offsets.into()) }; |
Contributor
There was a problem hiding this comment.
Here it would also be nice to comment about why this is safe (what assumptions it relies on). However, I see the existing code doesn't have a safety comment
// Safety: offsets are correctly constructed| predicate: &FilterPredicate, | ||
| ) -> GenericByteViewArray<T> { | ||
| let new_view_buffer = filter_native(array.views(), predicate); | ||
| let views = ScalarBuffer::new(new_view_buffer, 0, predicate.count); |
Contributor
There was a problem hiding this comment.
Here (and other places) you can probably use the unchecked variants too to skip some checks, if we need to get additional speed (ScalarBuffer::new_unchecked)
However, given your PR removes an allocation (the buffers array) I suspect this is already going to be faster and avoiding unsafe is a nice bonus ❤️
| } | ||
|
|
||
| /// Filters `nulls` and reuses the computed `null_count` to avoid scanning the bitmap. | ||
| fn filter_nulls(nulls: Option<&NullBuffer>, predicate: &FilterPredicate) -> Option<NullBuffer> { |
Contributor
There was a problem hiding this comment.
What do you think about making this a method on FilterPredicate? That would make it easier to find / reuse I think.
impl FilterPredicate {
fn filter_nulls(&self, nulls: Option<&NullBuffer>) -> Option<NullBuffer> {
...
}
}
|
🤖 Arrow criterion benchmark completed (GKE) | trigger Instance: CPU Details (lscpu)Details
Resource Usagebase (merge-base)
branch
File an issue against this benchmark runner |
Contributor
|
@liamzwbao -- some of these results look great but some look like they got slightly slower. Can you investigate (perhaps by using |
Contributor
|
run benchmark filter_kernels |
|
🤖 Arrow criterion benchmark running (GKE) | trigger CPU Details (lscpu)Comparing issue-9298-repalce-array-data-arrow-select (34b1837) to accb1cf (merge-base) diff File an issue against this benchmark runner |
|
🤖 Arrow criterion benchmark completed (GKE) | trigger Instance: CPU Details (lscpu)Details
Resource Usagebase (merge-base)
branch
File an issue against this benchmark runner |
Contributor
|
🤔 hmm that run #9986 (comment) looks very good. Will run another |
Contributor
|
run benchmark filter_kernels |
|
🤖 Arrow criterion benchmark running (GKE) | trigger CPU Details (lscpu)Comparing issue-9298-repalce-array-data-arrow-select (34b1837) to accb1cf (merge-base) diff File an issue against this benchmark runner |
|
🤖 Arrow criterion benchmark completed (GKE) | trigger Instance: CPU Details (lscpu)Details
Resource Usagebase (merge-base)
branch
File an issue against this benchmark runner |
Contributor
|
I would say the benchmarks don't show a clear win / downside one way or the other I do think it would be good to avoid the new checks just to be sure but otherwise this PR is ready to go. Thank you again |
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
3 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.
Which issue does this PR close?
ArrayDatawith direct Array construction, when possible #9298.Rationale for this change
What changes are included in this PR?
ArrayDataBuilderpaths inarrow-select/src/filter.rswith direct typed array constructors.Are these changes tested?
Covered by exsiting tests
Are there any user-facing changes?
No