refactor[fsst]: take ArrayRef in fsst_compress, decide offset width upfront#7900
Merged
Conversation
github-actions
Bot
removed
the
action/benchmark
Merging this PR will improve performance by 16.33%
|
| Mode | Benchmark | BASE |
HEAD |
Efficiency | |
|---|---|---|---|---|---|
| ❌ | Simulation | chunked_varbinview_into_canonical[(1000, 10)] |
169 µs | 205.9 µs | -17.93% |
| ⚡ | Simulation | chunked_bool_canonical_into[(1000, 10)] |
27 µs | 16.3 µs | +66.21% |
| ⚡ | Simulation | eq_pushdown_low_match |
968.7 µs | 786 µs | +23.24% |
| ⚡ | Simulation | chunked_varbinview_opt_canonical_into[(1000, 10)] |
206.1 µs | 169.6 µs | +21.5% |
| ⚡ | Simulation | eq_pushdown_high_match |
1,063.9 µs | 880.5 µs | +20.83% |
| ⚡ | Simulation | chunked_varbinview_opt_into_canonical[(1000, 10)] |
219.4 µs | 183 µs | +19.95% |
| ⚡ | Simulation | bitwise_not_vortex_buffer_mut[128] |
244.4 ns | 215.3 ns | +13.55% |
| ⚡ | Simulation | pushdown_compare[(10000, 16, 8)] |
264.8 µs | 239.3 µs | +10.64% |
| ⚡ | Simulation | bitwise_not_vortex_buffer_mut[1024] |
304.7 ns | 275.6 ns | +10.58% |
| ⚡ | Simulation | eq_i64_constant |
318.9 µs | 289.1 µs | +10.32% |
| 🆕 | Simulation | eq_pushdown_high_match_view |
N/A | 882.1 µs | N/A |
| 🆕 | Simulation | eq_pushdown_low_match_view |
N/A | 779.4 µs | N/A |
Tip
Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.
Comparing mp/fsst-compress-refactor (a4b73a5) with develop (d53301e)
Contributor
Polar Signals Profiling ResultsLatest Run
Powered by Polar Signals Cloud |
Contributor
Benchmarks: PolarSignals ProfilingVortex (geomean): 0.895x ✅ datafusion / vortex-file-compressed (0.895x ✅, 5↑ 0↓)
|
Contributor
File Sizes: PolarSignals ProfilingNo file size changes detected. |
Contributor
Benchmarks: FineWeb NVMeVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (0.975x ➖, 1↑ 0↓)
datafusion / vortex-compact (0.983x ➖, 1↑ 1↓)
datafusion / parquet (0.939x ➖, 1↑ 0↓)
duckdb / vortex-file-compressed (0.987x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.011x ➖, 0↑ 1↓)
duckdb / parquet (0.951x ➖, 1↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: FineWeb NVMeNo file size changes detected. |
Contributor
Benchmarks: TPC-H SF=1 on NVMEVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (1.002x ➖, 0↑ 0↓)
datafusion / vortex-compact (0.994x ➖, 0↑ 0↓)
datafusion / parquet (1.003x ➖, 0↑ 1↓)
datafusion / arrow (1.000x ➖, 0↑ 1↓)
duckdb / vortex-file-compressed (0.997x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.990x ➖, 0↑ 0↓)
duckdb / parquet (0.999x ➖, 0↑ 1↓)
duckdb / duckdb (0.997x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: TPC-H SF=1 on NVMENo file size changes detected. |
Contributor
|
What happens if the compress actually increase the array code size this estimate might still be incorrect |
Contributor
Benchmarks: TPC-DS SF=1 on NVMEVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (1.010x ➖, 1↑ 1↓)
datafusion / vortex-compact (1.011x ➖, 0↑ 0↓)
datafusion / parquet (1.012x ➖, 1↑ 4↓)
duckdb / vortex-file-compressed (1.013x ➖, 1↑ 0↓)
duckdb / vortex-compact (1.010x ➖, 0↑ 1↓)
duckdb / parquet (1.009x ➖, 0↑ 0↓)
duckdb / duckdb (1.001x ➖, 3↑ 2↓)
Full attributed analysis
|
Contributor
File Sizes: TPC-DS SF=1 on NVMENo file size changes detected. |
Contributor
Benchmarks: FineWeb S3Verdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (0.942x ➖, 1↑ 0↓)
datafusion / vortex-compact (1.142x ➖, 0↑ 0↓)
datafusion / parquet (1.078x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (1.033x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.031x ➖, 0↑ 0↓)
duckdb / parquet (1.056x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
Benchmarks: Random AccessVortex (geomean): 0.964x ➖ unknown / unknown (0.980x ➖, 2↑ 0↓)
|
Contributor
Benchmarks: Statistical and Population GeneticsVerdict: No clear signal (low confidence) duckdb / vortex-file-compressed (1.009x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.999x ➖, 0↑ 0↓)
duckdb / parquet (0.987x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: Statistical and Population GeneticsNo file size changes detected. |
Contributor
Benchmarks: TPC-H SF=10 on NVMEVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (1.004x ➖, 0↑ 0↓)
datafusion / vortex-compact (1.002x ➖, 0↑ 0↓)
datafusion / parquet (1.007x ➖, 0↑ 0↓)
datafusion / arrow (1.007x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (1.000x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.999x ➖, 0↑ 0↓)
duckdb / parquet (1.008x ➖, 0↑ 0↓)
duckdb / duckdb (0.996x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: TPC-H SF=10 on NVMENo file size changes detected. |
Contributor
Benchmarks: Clickbench on NVMEVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (0.968x ➖, 0↑ 1↓)
datafusion / parquet (0.939x ➖, 2↑ 0↓)
duckdb / vortex-file-compressed (0.986x ➖, 2↑ 6↓)
duckdb / parquet (0.967x ➖, 1↑ 0↓)
duckdb / duckdb (0.950x ➖, 5↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: Clickbench on NVMEFile Size Changes (1 files changed, -0.0% overall, 0↑ 1↓)
Totals:
|
Contributor
Benchmarks: TPC-H SF=1 on S3Verdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (1.134x ➖, 0↑ 4↓)
datafusion / vortex-compact (1.044x ➖, 0↑ 1↓)
datafusion / parquet (1.049x ➖, 1↑ 2↓)
duckdb / vortex-file-compressed (1.033x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.084x ➖, 0↑ 0↓)
duckdb / parquet (1.026x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
Benchmarks: CompressionVortex (geomean): 1.008x ➖ unknown / unknown (1.004x ➖, 0↑ 3↓)
|
Contributor
Benchmarks: TPC-H SF=10 on S3Verdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (1.068x ➖, 0↑ 2↓)
datafusion / vortex-compact (1.056x ➖, 0↑ 1↓)
datafusion / parquet (0.994x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (1.062x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.051x ➖, 0↑ 0↓)
duckdb / parquet (1.087x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
|
the upper bound of fsst is len*2+7, we can choose the type that fits this value and that WILL always work, so anything in 2**31-4 length will be u32 and anything above will be u64 |
Contributor
|
I think you still want a compress method that can take ArrayRef so callers don't have to know what they're compressing and we don't have to expose internal of how fsst does compression, then internally you can switch on the encoding type. |
Member
|
was there a merge that went wrong? There seems to be a bunch of changes in here that are not relevant to this PR |
Contributor
Author
|
re-merged develop |
Contributor
|
This PR has been marked as stale because it has been open for 14 days with no activity. Please comment or remove the stale label if you wish to keep it active, otherwise it will be closed in 7 days |
robert3005
force-pushed
the
mp/fsst-compress-refactor
branch
from
3dec60f to
ceb5796
Compare
robert3005
changed the title
Contributor
|
Rebased onto develop as a single clean commit (the earlier merge noise is gone — the diff is now 23 fsst-related files). Status of review feedback:
The rebase also folds in develop-side changes: the #8324 test optimization (empty symbol table, ~half the memory) now runs through the new API, |
github-actions
Bot
removed
the
stale
…pfront fsst_compress and fsst_train_compressor take an ArrayRef and dispatch internally on VarBinView/VarBin encoding, so callers don't need to know which string encoding they hold. The codes-offsets width (i32 vs i64) is chosen upfront from the FSST worst-case bound (2 * input bytes + 7 per non-null row), which always covers the actual output. Follow-up to #7853 addressing review feedback on #7833. Signed-off-by: mprammer <martin@spiraldb.com> Signed-off-by: Robert Kruszewski <github@robertk.io> Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
robert3005
force-pushed
the
mp/fsst-compress-refactor
branch
from
e626e80 to
330f68d
Compare
Signed-off-by: Robert Kruszewski <github@robertk.io>
robert3005
approved these changes
Jun 22, 2026
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
4 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
Refactors
fsst_compressandfsst_train_compressorto take anArrayRefand dispatch internally on the input encoding (VarBinVieworVarBin), so callers don't need to know which string encoding they hold. The codes-offsets width (i32 vs i64) is decided upfront from the FSST worst-case bound (2 × input bytes + 7 per non-null row), which always covers the actual output regardless of how well the data compresses.Follow-up to #7853 addressing review feedback on #7833.
🤖 Generated with Claude Code