perf: don't use fill_null for executing an array into a mask#8466
Merged
Conversation
…sion) #8121 inserted `array.fill_null(false)?` before every `execute::<Mask>` filter call site. `fill_null` is a no-op only for non-nullable arrays; for nullable predicate results (e.g. `LIKE`/`<>` over nullable columns, as ClickBench string columns are stored) it takes the expensive path: `precondition` calls `array.validity()?` on the lazy predicate array, re-deriving validity from the expression tree and materializing an intermediate array, before `execute::<Mask>` canonicalizes again. This regressed ClickBench Q22. Add `execute_mask_coercing_nulls`, which canonicalizes the array once and folds validity into the value bits with a single bitmap AND (the pre-#8121 cost profile), and route the filter readers (dict/flat/partitioned/row_idx/zone_map, CUDA, is_constant) through it. `Mask::execute` stays strict about nullable input. Signed-off-by: Joe Isaacs <joe.isaacs@live.co.uk> Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
joseph-isaacs
added
the
action/benchmark-sql
github-actions
Bot
removed
the
action/benchmark-sql
Merging this PR will not alter performance
|
| Mode | Benchmark | BASE |
HEAD |
Efficiency | |
|---|---|---|---|---|---|
| ❌ | Simulation | chunked_varbinview_canonical_into[(1000, 10)] |
162.2 µs | 198.2 µs | -18.16% |
| ❌ | Simulation | chunked_varbinview_canonical_into[(100, 100)] |
273.2 µs | 308.6 µs | -11.49% |
| ❌ | Simulation | chunked_varbinview_into_canonical[(100, 100)] |
330.8 µs | 367.7 µs | -10.04% |
| ⚡ | Simulation | chunked_varbinview_opt_canonical_into[(1000, 10)] |
213.8 µs | 176.9 µs | +20.86% |
| ⚡ | Simulation | chunked_varbinview_opt_into_canonical[(1000, 10)] |
229.1 µs | 193.4 µs | +18.46% |
Tip
Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.
Comparing ji/fix-mask-fill-null-regression (b3c69d8) with develop (a4476f1)
Footnotes
-
3 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩
joseph-isaacs
added
the
action/benchmark
github-actions
Bot
removed
the
action/benchmark
Contributor
Polar Signals Profiling ResultsLatest Run
Powered by Polar Signals Cloud |
Contributor
Benchmarks: PolarSignals ProfilingVortex (geomean): 1.019x ➖ How to read Verdict and Engines
datafusion / vortex-file-compressed (1.019x ➖, 0↑ 1↓)
No file size changes detected. |
Contributor
Benchmarks: FineWeb NVMeVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.036x ➖, 0↑ 0↓)
datafusion / vortex-compact (0.987x ➖, 1↑ 1↓)
datafusion / parquet (1.005x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (0.998x ➖, 1↑ 2↓)
duckdb / vortex-compact (1.012x ➖, 0↑ 0↓)
duckdb / parquet (1.059x ➖, 0↑ 2↓)
File Size Changes (1 files changed, +0.0% overall, 1↑ 0↓)
Totals:
|
Contributor
Benchmarks: TPC-H SF=1 on NVMEVerdict: No clear signal (environment too noisy confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.004x ➖, 0↑ 0↓)
datafusion / vortex-compact (1.019x ➖, 0↑ 0↓)
datafusion / parquet (1.005x ➖, 0↑ 1↓)
datafusion / arrow (1.004x ➖, 1↑ 0↓)
duckdb / vortex-file-compressed (1.007x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.005x ➖, 0↑ 0↓)
duckdb / parquet (1.007x ➖, 1↑ 2↓)
duckdb / duckdb (1.013x ➖, 0↑ 0↓)
File Size Changes (10 files changed, -0.0% overall, 3↑ 7↓)
Totals:
|
Contributor
Benchmarks: TPC-DS SF=1 on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.021x ➖, 0↑ 4↓)
datafusion / vortex-compact (1.020x ➖, 2↑ 4↓)
datafusion / parquet (1.022x ➖, 1↑ 4↓)
duckdb / vortex-file-compressed (1.014x ➖, 0↑ 2↓)
duckdb / vortex-compact (1.013x ➖, 2↑ 3↓)
duckdb / parquet (1.006x ➖, 1↑ 1↓)
duckdb / duckdb (1.001x ➖, 1↑ 2↓)
File Size Changes (6 files changed, +0.0% overall, 2↑ 4↓)
Totals:
|
Contributor
Benchmarks: FineWeb S3Verdict: No clear signal (environment too noisy confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.282x ➖, 0↑ 3↓)
datafusion / vortex-compact (0.869x ➖, 1↑ 0↓)
datafusion / parquet (1.002x ➖, 1↑ 0↓)
duckdb / vortex-file-compressed (0.989x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.938x ➖, 1↑ 0↓)
duckdb / parquet (0.961x ➖, 0↑ 0↓)
|
Contributor
Benchmarks: Statistical and Population GeneticsVerdict: No clear signal (low confidence) How to read Verdict and Engines
duckdb / vortex-file-compressed (1.007x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.998x ➖, 0↑ 0↓)
duckdb / parquet (0.998x ➖, 0↑ 0↓)
File Size Changes (1 files changed, +0.0% overall, 1↑ 0↓)
Totals:
|
Contributor
Benchmarks: TPC-H SF=10 on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.160x ❌, 0↑ 19↓)
datafusion / vortex-compact (1.136x ❌, 0↑ 18↓)
datafusion / parquet (1.064x ➖, 0↑ 4↓)
datafusion / arrow (1.093x ➖, 0↑ 11↓)
duckdb / vortex-file-compressed (1.123x ❌, 0↑ 17↓)
duckdb / vortex-compact (1.102x ❌, 0↑ 13↓)
duckdb / parquet (1.066x ➖, 0↑ 3↓)
duckdb / duckdb (1.088x ➖, 0↑ 9↓)
File Size Changes (26 files changed, -0.0% overall, 13↑ 13↓)
Totals:
|
Contributor
Benchmarks: Clickbench on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.949x ➖, 6↑ 1↓)
datafusion / parquet (0.977x ➖, 2↑ 1↓)
duckdb / vortex-file-compressed (0.968x ➖, 2↑ 0↓)
duckdb / parquet (0.997x ➖, 0↑ 0↓)
duckdb / duckdb (0.956x ➖, 6↑ 0↓)
File Size Changes (105 files changed, -0.0% overall, 48↑ 57↓)
Totals:
|
Contributor
Benchmarks: TPC-H SF=1 on S3Verdict: No clear signal (environment too noisy confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.858x ➖, 8↑ 4↓)
datafusion / vortex-compact (1.058x ➖, 0↑ 3↓)
datafusion / parquet (1.096x ➖, 0↑ 4↓)
duckdb / vortex-file-compressed (1.082x ➖, 0↑ 1↓)
duckdb / vortex-compact (1.044x ➖, 0↑ 1↓)
duckdb / parquet (1.033x ➖, 0↑ 0↓)
|
Contributor
Benchmarks: Appian on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.141x ❌, 0↑ 6↓)
datafusion / parquet (1.018x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (1.086x ➖, 0↑ 4↓)
duckdb / parquet (1.075x ➖, 0↑ 0↓)
duckdb / duckdb (1.076x ➖, 0↑ 2↓)
File Size Changes (4 files changed, -0.0% overall, 0↑ 4↓)
Totals:
|
Contributor
Benchmarks: TPC-H SF=10 on S3Verdict: No clear signal (environment too noisy confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.072x ➖, 0↑ 4↓)
datafusion / vortex-compact (1.069x ➖, 1↑ 2↓)
datafusion / parquet (0.979x ➖, 1↑ 3↓)
duckdb / vortex-file-compressed (1.017x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.023x ➖, 0↑ 0↓)
duckdb / parquet (0.987x ➖, 0↑ 0↓)
|
Replace the execute_mask_coercing_nulls helper with an Executor-style `ArrayRef::null_as_false()` adapter (matching the design in #8121's history): canonicalize once and fold validity into the value bits with a single AND that reuses the value buffer when uniquely owned, deferring non-nullable input to the strict `Mask::execute`. Add the mask_fill_null microbenchmark and a one-line note that `fill_null` on a lazy `ScalarFn` array is currently slow. Verified on CI ClickBench: clickbench_q22/duckdb:vortex 1126ms -> 580ms (0.51, ~2x). Signed-off-by: Joe Isaacs <joe.isaacs@live.co.uk> Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
joseph-isaacs
added
the
action/benchmark-sql
github-actions
Bot
removed
the
action/benchmark-sql
joseph-isaacs
added
the
action/benchmark
github-actions
Bot
removed
the
action/benchmark
Signed-off-by: Joe Isaacs <joe.isaacs@live.co.uk> Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
joseph-isaacs
changed the title
onursatici
approved these changes
Jun 17, 2026
Clippy redundant_clone fired on array.clone() in null_as_false_matches_fill_null_then_mask since fill_null borrows self and the array is reused afterwards. Signed-off-by: Joe Isaacs <joe.isaacs@live.co.uk>
joseph-isaacs
added
the
changelog/performance
Claude
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.
Our current validity for a ScalarFnArray is very bad. Instead execute and merge buffers for now. See clickbench:duckdb:q22
See #8471