perf: Optimize NULL handling in `StringViewArrayBuilder` by neilconway · Pull Request #21538 · apache/datafusion

neilconway · 2026-04-10T16:12:44Z

Which issue does this PR close?

Closes Optimize NULL handling in StringViewArrayBuilder #21537.

Rationale for this change

StringViewArrayBuilder is implemented on top of Arrow's StringViewBuilder; the latter tracks NULLs incrementally. However, the StringViewArrayBuilder requires callers to pass a NULL buffer to finish() anyway, so the NULL bitmap that has been computed by StringViewBuilder is discarded. It would be more efficient to stop using StringViewBuilder so that we don't do this redundant work; in theory there might be room for inconsistency between the two NULL bitmaps as well.

Right now, StringViewArrayBuilder is only used by the concat and concat_ws UDFs, but I'd like to generalize the API and use it more broadly in place of StringViewBuilder (#21539). For the time being, here are the results of this PR on the concat benchmarks (Arm64):

  - 1024 rows: 29.6 µs → 28.0 µs, -5.3%
  - 4096 rows: 134.3 µs → 125.6 µs, -6.5%
  - 8192 rows: 289.7 µs → 273.5 µs, -5.6%

What changes are included in this PR?

Stop using StringViewBuilder and build the views ourselves
Improve some comments
Return an error instead of panicking on large input strings

Are these changes tested?

Yes.

Are there any user-facing changes?

No.

neilconway · 2026-04-10T16:14:55Z

FYI @Omega359

Omega359

I looked this over and it seems reasonable, thanks!

neilconway · 2026-04-13T15:39:58Z

@martin-g @timsaucer Thanks for the reviews! I noticed that the existing code panicked for large input strings, so I adjusted that to return an error (via Result) instead.

alamb · 2026-04-13T17:28:25Z

run benchmark concat

adriangbot · 2026-04-13T17:30:29Z

🤖 Criterion benchmark running (GKE) | trigger
Instance: c4a-highmem-16 (12 vCPU / 65 GiB) | Linux bench-c4238361261-1168-r48dj 6.12.55+ #1 SMP Sun Feb 1 08:59:41 UTC 2026 aarch64 GNU/Linux

CPU Details (lscpu)

Architecture:                            aarch64
CPU op-mode(s):                          64-bit
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               ARM
Model name:                              Neoverse-V2
Model:                                   1
Thread(s) per core:                      1
Core(s) per cluster:                     16
Socket(s):                               -
Cluster(s):                              1
Stepping:                                r0p1
BogoMIPS:                                2000.00
Flags:                                   fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm jscvt fcma lrcpc dcpop sha3 sm3 sm4 asimddp sha512 sve asimdfhm dit uscat ilrcpc flagm sb paca pacg dcpodp sve2 sveaes svepmull svebitperm svesha3 svesm4 flagm2 frint svei8mm svebf16 i8mm bf16 dgh rng bti
L1d cache:                               1 MiB (16 instances)
L1i cache:                               1 MiB (16 instances)
L2 cache:                                32 MiB (16 instances)
L3 cache:                                80 MiB (1 instance)
NUMA node(s):                            1
NUMA node0 CPU(s):                       0-15
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Not affected
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Not affected
Vulnerability Spec rstack overflow:      Not affected
Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:                Mitigation; __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; CSV2, BHB
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Comparing neilc/perf-string-view-builder-nulls (4e33f61) to 4389f14 (merge-base) diff
BENCH_NAME=concat
BENCH_COMMAND=cargo bench --features=parquet --bench concat
BENCH_FILTER=
Results will be posted here when complete

File an issue against this benchmark runner

adriangbot · 2026-04-13T17:34:27Z

🤖 Criterion benchmark completed (GKE) | trigger

Instance: c4a-highmem-16 (12 vCPU / 65 GiB)

CPU Details (lscpu)

Architecture:                            aarch64
CPU op-mode(s):                          64-bit
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               ARM
Model name:                              Neoverse-V2
Model:                                   1
Thread(s) per core:                      1
Core(s) per cluster:                     16
Socket(s):                               -
Cluster(s):                              1
Stepping:                                r0p1
BogoMIPS:                                2000.00
Flags:                                   fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm jscvt fcma lrcpc dcpop sha3 sm3 sm4 asimddp sha512 sve asimdfhm dit uscat ilrcpc flagm sb paca pacg dcpodp sve2 sveaes svepmull svebitperm svesha3 svesm4 flagm2 frint svei8mm svebf16 i8mm bf16 dgh rng bti
L1d cache:                               1 MiB (16 instances)
L1i cache:                               1 MiB (16 instances)
L2 cache:                                32 MiB (16 instances)
L3 cache:                                80 MiB (1 instance)
NUMA node(s):                            1
NUMA node0 CPU(s):                       0-15
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Not affected
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Not affected
Vulnerability Spec rstack overflow:      Not affected
Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:                Mitigation; __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; CSV2, BHB
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Details

group                               main                                   neilc_perf-string-view-builder-nulls
-----                               ----                                   ------------------------------------
concat function/concat/1024         1.00     15.5±0.43µs        ? ?/sec    1.00     15.6±0.42µs        ? ?/sec
concat function/concat/4096         1.00     62.4±1.73µs        ? ?/sec    1.01     62.9±1.63µs        ? ?/sec
concat function/concat/8192         1.00    124.3±3.55µs        ? ?/sec    1.00    123.8±3.78µs        ? ?/sec
concat function/concat/scalar       1.00   1013.8±5.36ns        ? ?/sec    1.01   1020.6±3.52ns        ? ?/sec
concat function/concat_view/1024    1.03     40.6±0.19µs        ? ?/sec    1.00     39.6±0.07µs        ? ?/sec
concat function/concat_view/4096    1.02    199.0±0.43µs        ? ?/sec    1.00    194.6±0.48µs        ? ?/sec
concat function/concat_view/8192    1.01    422.3±0.55µs        ? ?/sec    1.00    416.3±0.72µs        ? ?/sec

Resource Usage

base (merge-base)

Metric	Value
Wall time	69.1s
Peak memory	3.5 GiB
Avg memory	3.5 GiB
CPU user	85.6s
CPU sys	0.9s
Peak spill	0 B

branch

Metric	Value
Wall time	68.7s
Peak memory	3.5 GiB
Avg memory	3.5 GiB
CPU user	86.1s
CPU sys	0.2s
Peak spill	0 B

File an issue against this benchmark runner

neilconway · 2026-04-14T13:52:02Z

@martin-g Is this okay to merge?

neilconway · 2026-04-14T14:27:57Z

Thanks @martin-g @timsaucer !

alamb · 2026-04-13T17:28:08Z

    }

-    pub fn append_offset(&mut self) {
+    pub fn append_offset(&mut self) -> Result<()> {


I think this is an API change we should call out (I'll add a api-change label)

Is there a reason why StringViewArrayBuilder is pub, rather than pub(crate)? It is a fairly special-purpose API.

TLDR is I don't know -- it got added for some very special purpose reason I can't remember (I think ideally we would use the upstream one in arrow-rs, but for some reason we needed a special copy. I can't remember why)

I wrote out a quick plan for what I'd like to do here, see #21684 ; comments very welcome

## Which issue does this PR close? - Closes apache#21537. ## Rationale for this change `StringViewArrayBuilder` is implemented on top of Arrow's `StringViewBuilder`; the latter tracks NULLs incrementally. However, the `StringViewArrayBuilder` requires callers to pass a NULL buffer to `finish()` anyway, so the NULL bitmap that has been computed by `StringViewBuilder` is discarded. It would be more efficient to stop using `StringViewBuilder` so that we don't do this redundant work; in theory there might be room for inconsistency between the two NULL bitmaps as well. Right now, `StringViewArrayBuilder` is only used by the `concat` and `concat_ws` UDFs, but I'd like to generalize the API and use it more broadly in place of `StringViewBuilder` (apache#21539). For the time being, here are the results of this PR on the `concat` benchmarks (Arm64): ``` - 1024 rows: 29.6 µs → 28.0 µs, -5.3% - 4096 rows: 134.3 µs → 125.6 µs, -6.5% - 8192 rows: 289.7 µs → 273.5 µs, -5.6% ``` ## What changes are included in this PR? * Stop using `StringViewBuilder` and build the views ourselves * Improve some comments * Return an error instead of panicking on large input strings ## Are these changes tested? Yes. ## Are there any user-facing changes? No. --------- Co-authored-by: Martin Grigorov <martin-g@users.noreply.github.com>

…e#21695) ## Which issue does this PR close? - Part of apache#21684 but does not close it. ## Rationale for this change Rename the three builders in `datafusion/functions/src/strings.rs` to make their special-purpose nature explicit: - `StringArrayBuilder` -> `ConcatStringBuilder` - `LargeStringArrayBuilder` -> `ConcatLargeStringBuilder` - `StringViewArrayBuilder` -> `ConcatStringViewBuilder` These builders are used only by `concat` and `concat_ws`, and their APIs are specific enough to the needs of `concat`-like callers that it seems unlikely that other call-sites will emerge in the future. This also frees the previous names of these builders to be used by a more broadly useful string builder API. Also make these types `pub(crate)`, instead of `pub`; there is no good reason to make them part of the public API. This is a breaking API change. ## What changes are included in this PR? * Rename types * Make types `pub(crate)`, not `pub` * Add section to v54 migration guide ## Are these changes tested? Yes. ## Are there any user-facing changes? Yes, breaking API change -- although these types should probably not have been part of the public API to begin with, and we already made a breaking API change to them as part of other work in v54 (apache#21538)

.

ccbb95f

github-actions Bot added the functions Changes to functions implementation label Apr 10, 2026

Tweak comment

b17be59

mbutrovich mentioned this pull request Apr 10, 2026

feat: Add immediate mode option for native shuffle apache/datafusion-comet#3845

Draft

Omega359 approved these changes Apr 10, 2026

View reviewed changes

martin-g reviewed Apr 13, 2026

View reviewed changes

Comment thread datafusion/functions/src/strings.rs Outdated

Comment thread datafusion/functions/src/strings.rs Outdated

Comment thread datafusion/functions/src/strings.rs Outdated

timsaucer approved these changes Apr 13, 2026

View reviewed changes

neilconway added 2 commits April 13, 2026 10:51

Improvements per review comments

b12dbd7

cargo fmt, fix error handling

4e33f61

alamb added performance Make DataFusion faster api change Changes the API exposed to users of the crate labels Apr 13, 2026

martin-g approved these changes Apr 13, 2026

View reviewed changes

Merge branch 'main' into neilc/perf-string-view-builder-nulls

3a8cdf7

martin-g enabled auto-merge April 14, 2026 13:54

martin-g added this pull request to the merge queue Apr 14, 2026

Merged via the queue into apache:main with commit 961c5fc Apr 14, 2026
31 checks passed

alamb reviewed Apr 14, 2026

View reviewed changes

neilconway deleted the neilc/perf-string-view-builder-nulls branch April 16, 2026 16:30

neilconway mentioned this pull request Apr 17, 2026

chore: Rename concat-specific string builders, make pub(crate) #21695

Merged

Conversation

neilconway commented Apr 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Which issue does this PR close?

Rationale for this change

What changes are included in this PR?

Are these changes tested?

Are there any user-facing changes?

Uh oh!

neilconway commented Apr 10, 2026

Uh oh!

Omega359 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

neilconway commented Apr 13, 2026

Uh oh!

alamb commented Apr 13, 2026

Uh oh!

adriangbot commented Apr 13, 2026

Uh oh!

adriangbot commented Apr 13, 2026

Uh oh!

neilconway commented Apr 14, 2026

Uh oh!

Uh oh!

neilconway commented Apr 14, 2026

Uh oh!

alamb Apr 13, 2026

Choose a reason for hiding this comment

Uh oh!

neilconway Apr 16, 2026

Choose a reason for hiding this comment

Uh oh!

alamb Apr 16, 2026

Choose a reason for hiding this comment

Uh oh!

neilconway Apr 16, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

neilconway commented Apr 10, 2026 •

edited

Loading