Skip to content

Tune V2 scan tasks for sparse filters and unfiltered projections - #10969

Open
joseph-isaacs wants to merge 2 commits into
developfrom
ji/codex/clickbench-pruned-parallelism
Open

joseph-isaacs wants to merge 2 commits into
developfrom
ji/codex/clickbench-pruned-parallelism

Conversation

@joseph-isaacs

@joseph-isaacs joseph-isaacs commented Oct 29, 2026 •

Copy link
Copy Markdown
Contributor

V2 spends too much time setting up small unfiltered tasks, and can leave workers idle after heavy zone pruning. Size work chunks for those two cases while retaining bounded output batches:

  • Group unfiltered, non-identity projections into 256K-row tasks. Each task still emits batches of at most 64K rows.
  • Subdivide general 128K filtered tasks into 32K tasks when pruning removes at least half of their rows. Keep the larger tasks used for cheap single-column filters.

This follows merged #10799. Q38 returns the same 47,740 rows as V1 with fewer bytes read and about 26% fewer sampled CPU cycles, but the original V2 path gathered its output into only eight batches versus V1's 53. Its three surviving ranges retain 13–42% of their rows. Q25 usually retains over 90%, so applying smaller tasks to every partially pruned range adds overhead.

Performance

The unfiltered diagnostic returns the same approximately 100M rows in 1,620 output batches for Q2/Q3/Q29. Q3's read requests fall from 543 to 372 while bytes stay approximately 331 MB: fewer tasks also let more reads coalesce.

Matched local ARM64 DataFusion/Vortex runs used CPUs 2–31 and matching Tokio worker/DataFusion partition counts, without detected competing benchmark or build overlap. The following compares the existing scheduling policy against the relevant isolated change:

Query / change Before, 30 workers After, 30 workers
Q2, 256K unfiltered tasks 12.64 ms 11.86 ms (-6.1%)
Q3, 256K unfiltered tasks 14.92 ms 13.69 ms (-8.3%)
Q29, 256K unfiltered tasks 21.58 ms 20.51 ms (-5.0%)
Q38, 32K heavily pruned tasks 7.85 ms 6.60 ms (-15.9%)

Q2/Q3/Q29 use 50 warm samples, with the baseline repeated before and after the candidates. Q38 uses 20 warm samples on the rebased source; earlier counterbalanced 100-sample runs reproduced its improvement. At 8 workers, the longer unfiltered repeats improved Q2/Q3/Q29 by approximately 2.9%/5.4%/1.2%.

A full 43-query comparison at 30 workers and 17 unfiltered queries at 8 workers completed. The latter had a 0.987 geometric runtime ratio. Some controls in the full run were noisy, including Q2, so no whole-suite speedup is claimed. The filtered subdivision has a tradeoff on smaller machines: Q40/Q41 added about 0.17–0.18 ms at 8 workers.

The first CI benchmark, which includes only the filtered subdivision, confirms Q38 13.73 → 12.19 ms (-11.3%). Its DataFusion/Vortex geometric ratio is 0.994; attribution reports no clear suite-wide signal. Q40/Q41 were within 1%, while Q42 was +7.3%, which did not reproduce locally. Fresh CI benchmarks are requested for the combined change.

The local measurements used a verified diagnostic executable based on 2eae76464d plus this policy, with switches isolating task sizes. The final source is rebased onto develop e4dce8b74e; the benchmark switches are not part of the PR. For a source-to-source repeat, build the base and PR with identical settings:

cargo +1.98.0 rustc --locked -p datafusion-bench --bin datafusion-bench \
  --profile release_debug -j 8 -- -C strip=debuginfo
TOKIO_WORKER_THREADS=30 DATAFUSION_EXECUTION_TARGET_PARTITIONS=30 RUST_LOG=warn \
  taskset -c 2-31 <binary> clickbench --formats vortex --scan-v2 true \
  --iterations 51 --threads 30 --queries 2,3,29,36,37,38,39,40,41,42 \
  --display-format gh-json --hide-progress-bar \
  --opt remote-data-dir=file:///mnt/vortex-ssd/votex-4/data/clickbench_partitioned/

Discard the first iteration and compare warm medians.

Validation

  • 110 V2 scan tests passed on the pushed revision, rebased onto develop e4dce8b74e.
  • Targeted command: cargo +1.98.0 test --locked -p vortex-layout --lib --profile release_debug -j 8 scan::v2::.
  • Regression cases cover no pruning, all rows retained, mostly retained rows, the half-retained boundary, sparse rows, and cheap single-column scans.
  • The new unfiltered test checks that one 256K task emits four 64K batches and preserves all values.
  • Pinned nightly-2026-09-10 formatting and patch whitespace checks passed.
  • Linters and Tests, Rust Instrumented, Linux musl, Publish Dry Runs, CUDA, Codspeed, REUSE, typo, and web checks passed on 92221095.
  • Combined SQL benchmarks are running. TPC-H SF1 NVMe is neutral. PolarSignals warm performance is neutral (0.999x), but cold runtime is 1.194x versus cached base 2a9c7628; a repeat is pending. GitHub does not allow rerunning that job until the SQL workflow completes.

@joseph-isaacs joseph-isaacs added the action/bench-sql Run SQL benchmarks without Vortex Compact on this PR label Oct 29, 2026
@github-actions github-actions Bot removed the action/bench-sql Run SQL benchmarks without Vortex Compact on this PR label Oct 29, 2026
@joseph-isaacs joseph-isaacs added the changelog/performance A performance improvement label Oct 29, 2026 — with ChatGPT Codex Connector
@github-actions

github-actions Bot commented Oct 29, 2026 •

Copy link
Copy Markdown
Contributor

Polar Signals Profiling Results

Latest Run

Status Commit Job Attempt Link
🟢 Done 9222109 fineweb 1 Explore Profiling Data
🟢 Done 9222109 polarsignals 1 Explore Profiling Data
🟢 Done 9222109 tpch-nvme 1 Explore Profiling Data
🟢 Done 9222109 tpch-s3 1 Explore Profiling Data
Previous Runs (7)
Status Commit Job Attempt Link
🟢 Done a3d0ea3 fineweb 1 Explore Profiling Data
🟢 Done a3d0ea3 clickbench-nvme 1 Explore Profiling Data
🟢 Done a3d0ea3 polarsignals 1 Explore Profiling Data
🟢 Done a3d0ea3 tpch-s3 1 Explore Profiling Data
🟢 Done a3d0ea3 tpch-nvme-10 1 Explore Profiling Data
🟢 Done a3d0ea3 tpch-nvme 1 Explore Profiling Data
🟢 Done a3d0ea3 tpcds-nvme 1 Explore Profiling Data

Powered by Polar Signals Cloud

@github-actions

github-actions Bot commented Oct 29, 2026 •

Copy link
Copy Markdown
Contributor

Benchmarks: TPC-H SF=1 on NVME 📖

Commits: PR 92221095 vs base 6cc1b921
Verdict: No clear signal (low confidence)
Attributed Vortex impact: +0.2%
Engines: DataFusion No clear signal (-0.5%, low confidence) · DuckDB No clear signal (+0.8%, low confidence)
Vortex (geomean): hot 1.002x ➖ · cold 0.985x ➖
Parquet (geomean): hot 1.001x ➖ · cold 0.997x ➖
Shifts: Parquet (control) +0.1% · Median polish +0.1%

How to read Verdict and Engines
  • Verdict: Overall PR-level signal after subtracting baseline drift estimated from Parquet control rows. It can be Likely improvement, Likely regression, or No clear signal.
  • Engines: Per-engine attribution. DataFusion is compared against DataFusion/Parquet controls; DuckDB is compared against DuckDB/Parquet controls. This answers whether each engine improved or regressed independently.
  • Confidence: Based on directional consistency, share of rows above the noise floor, and control-run noise.
  • Hot vs cold: Every measurement is run several times. The first run is reported as the cold run, and the median of the runs after it is reported as the hot run. The verdict and significance use hot runs; each target's geomean reports hot and cold beside each other where the individual runs were recorded and hot alone where they were not, the cold column shows first-run cost per row, and hot/cold is how much of each run the warm path saves. Rows whose results predate per-run reporting show only one value, taken from the value the runner reported.
  • Table cells: Each cell reads PR / base / %diff. The hot and cold columns mark a change as 🔴 slower or 🟢 faster once it clears this suite's threshold; anything smaller is noise here and is left unmarked, as is hot/cold, because a shift in the warm-up ratio is not a win or a loss by itself.

datafusion / vortex-file-compressed / ns (0.996x ➖, 0↑ 0↓)
name hot ns (PR / base / %diff) cold ns (PR / base / %diff) hot/cold (PR / base / %diff)
tpch_q01/datafusion:vortex-file-compressed 15359271 / 15302474 / +0.4% 42862325 / 43418373 / -1.3% 0.36 / 0.35 / +1.7%
tpch_q02/datafusion:vortex-file-compressed 16688076 / 16660213 / +0.2% 36980743 / 35416377 / +4.4% 0.45 / 0.47 / -4.1%
tpch_q03/datafusion:vortex-file-compressed 19511354 / 19928511 / -2.1% 49015915 / 50509554 / -3.0% 0.40 / 0.39 / +0.9%
tpch_q04/datafusion:vortex-file-compressed 10380309 / 10552099 / -1.6% 35776820 / 35166301 / +1.7% 0.29 / 0.30 / -3.3%
tpch_q05/datafusion:vortex-file-compressed 37024433 / 37233191 / -0.6% 64294384 / 67470298 / -4.7% 0.58 / 0.55 / +4.4%
tpch_q06/datafusion:vortex-file-compressed 5784095 / 6158807 / -6.1% 29366092 / 29078331 / +1.0% 0.20 / 0.21 / -7.0%
tpch_q07/datafusion:vortex-file-compressed 51547366 / 52239674 / -1.3% 76773825 / 81989384 / -6.4% 0.67 / 0.64 / +5.4%
tpch_q08/datafusion:vortex-file-compressed 51084671 / 51118886 / -0.1% 85139198 / 79947840 / +6.5% 0.60 / 0.64 / -6.2%
tpch_q09/datafusion:vortex-file-compressed 47183412 / 47172231 / +0.0% 78109054 / 79925613 / -2.3% 0.60 / 0.59 / +2.3%
tpch_q10/datafusion:vortex-file-compressed 17616455 / 17219321 / +2.3% 52347547 / 54302681 / -3.6% 0.34 / 0.32 / +6.1%
tpch_q11/datafusion:vortex-file-compressed 14173798 / 14202214 / -0.2% 33811541 / 34343851 / -1.5% 0.42 / 0.41 / +1.4%
tpch_q12/datafusion:vortex-file-compressed 18211044 / 18168565 / +0.2% 44455466 / 44965720 / -1.1% 0.41 / 0.40 / +1.4%
tpch_q13/datafusion:vortex-file-compressed 19487794 / 19245801 / +1.3% 41186307 / 40946025 / +0.6% 0.47 / 0.47 / +0.7%
tpch_q14/datafusion:vortex-file-compressed 14359094 / 14518508 / -1.1% 39815676 / 42214760 / -5.7% 0.36 / 0.34 / +4.9%
tpch_q15/datafusion:vortex-file-compressed 18469255 / 18699425 / -1.2% 46444852 / 47491657 / -2.2% 0.40 / 0.39 / +1.0%
tpch_q16/datafusion:vortex-file-compressed 18446848 / 18404836 / +0.2% 41282588 / 40932366 / +0.9% 0.45 / 0.45 / -0.6%
tpch_q17/datafusion:vortex-file-compressed 27656221 / 28305141 / -2.3% 58712293 / 59162165 / -0.8% 0.47 / 0.48 / -1.5%
tpch_q18/datafusion:vortex-file-compressed 36013559 / 35784180 / +0.6% 60358094 / 58947758 / +2.4% 0.60 / 0.61 / -1.7%
tpch_q19/datafusion:vortex-file-compressed 12133920 / 12164596 / -0.3% 40007405 / 40378418 / -0.9% 0.30 / 0.30 / +0.7%
tpch_q20/datafusion:vortex-file-compressed 20170514 / 20324278 / -0.8% 46496343 / 46639287 / -0.3% 0.43 / 0.44 / -0.5%
tpch_q21/datafusion:vortex-file-compressed 39736637 / 39254123 / +1.2% 73097505 / 73449550 / -0.5% 0.54 / 0.53 / +1.7%
tpch_q22/datafusion:vortex-file-compressed 10570027 / 10398066 / +1.7% 30567223 / 31654820 / -3.4% 0.35 / 0.33 / +5.3%
datafusion / parquet / ns (1.000x ➖, 0↑ 0↓)
name hot ns (PR / base / %diff) cold ns (PR / base / %diff) hot/cold (PR / base / %diff)
tpch_q01/datafusion:parquet 71991330 / 72060429 / -0.1% 90533773 / 91087861 / -0.6% 0.80 / 0.79 / +0.5%
tpch_q02/datafusion:parquet 75728471 / 76250844 / -0.7% 94245654 / 94981310 / -0.8% 0.80 / 0.80 / +0.1%
tpch_q03/datafusion:parquet 83791951 / 84286940 / -0.6% 105105330 / 106915327 / -1.7% 0.80 / 0.79 / +1.1%
tpch_q04/datafusion:parquet 40749307 / 40260206 / +1.2% 62134509 / 59425365 / +4.6% 0.66 / 0.68 / -3.2%
tpch_q05/datafusion:parquet 112918781 / 112677227 / +0.2% 137989931 / 141906320 / -2.8% 0.82 / 0.79 / +3.1%
tpch_q06/datafusion:parquet 32252840 / 32013506 / +0.7% 49711177 / 52286997 / -4.9% 0.65 / 0.61 / +6.0%
tpch_q07/datafusion:parquet 115013869 / 114748813 / +0.2% 141930296 / 142208939 / -0.2% 0.81 / 0.81 / +0.4%
tpch_q08/datafusion:parquet 118749094 / 119965306 / -1.0% 147974863 / 149830569 / -1.2% 0.80 / 0.80 / +0.2%
tpch_q09/datafusion:parquet 140001547 / 140000897 / +0.0% 169953781 / 169247607 / +0.4% 0.82 / 0.83 / -0.4%
tpch_q10/datafusion:parquet 107046281 / 107235422 / -0.2% 131007321 / 131374057 / -0.3% 0.82 / 0.82 / +0.1%
tpch_q11/datafusion:parquet 61221214 / 61513511 / -0.5% 81000228 / 83022111 / -2.4% 0.76 / 0.74 / +2.0%
tpch_q12/datafusion:parquet 65798321 / 64560975 / +1.9% 88258313 / 84522197 / +4.4% 0.75 / 0.76 / -2.4%
tpch_q13/datafusion:parquet 147498054 / 147169076 / +0.2% 168417953 / 169678787 / -0.7% 0.88 / 0.87 / +1.0%
tpch_q14/datafusion:parquet 47395281 / 47683415 / -0.6% 74282995 / 74110982 / +0.2% 0.64 / 0.64 / -0.8%
tpch_q15/datafusion:parquet 74620485 / 75451616 / -1.1% 94215692 / 96438254 / -2.3% 0.79 / 0.78 / +1.2%
tpch_q16/datafusion:parquet 44144951 / 43574518 / +1.3% 60471834 / 60982845 / -0.8% 0.73 / 0.71 / +2.2%
tpch_q17/datafusion:parquet 124037885 / 124004328 / +0.0% 148482432 / 149206885 / -0.5% 0.84 / 0.83 / +0.5%
tpch_q18/datafusion:parquet 123781171 / 124026661 / -0.2% 145315546 / 144602509 / +0.5% 0.85 / 0.86 / -0.7%
tpch_q19/datafusion:parquet 57366306 / 57165426 / +0.4% 82018852 / 83572167 / -1.9% 0.70 / 0.68 / +2.3%
tpch_q20/datafusion:parquet 79583864 / 78868658 / +0.9% 104472097 / 107853795 / -3.1% 0.76 / 0.73 / +4.2%
tpch_q21/datafusion:parquet 116557660 / 116762701 / -0.2% 141328194 / 141654945 / -0.2% 0.82 / 0.82 / +0.1%
tpch_q22/datafusion:parquet 51592094 / 52522395 / -1.8% 69072300 / 68600321 / +0.7% 0.75 / 0.77 / -2.4%
duckdb / vortex-file-compressed / ns (1.009x ➖, 0↑ 1↓)
name hot ns (PR / base / %diff) cold ns (PR / base / %diff) hot/cold (PR / base / %diff)
tpch_q01/duckdb:vortex-file-compressed 9335283 / 7983105 / +16.9% 🔴 20369183 / 20188475 / +0.9% 0.46 / 0.40 / +15.9%
tpch_q02/duckdb:vortex-file-compressed 15653783 / 15517834 / +0.9% 25217747 / 26317936 / -4.2% 0.62 / 0.59 / +5.3%
tpch_q03/duckdb:vortex-file-compressed 18319361 / 18121943 / +1.1% 34181982 / 35169974 / -2.8% 0.54 / 0.52 / +4.0%
tpch_q04/duckdb:vortex-file-compressed 12073916 / 12464185 / -3.1% 25473193 / 26231075 / -2.9% 0.47 / 0.48 / -0.2%
tpch_q05/duckdb:vortex-file-compressed 20053729 / 20481417 / -2.1% 33803493 / 33771468 / +0.1% 0.59 / 0.61 / -2.2%
tpch_q06/duckdb:vortex-file-compressed 4738779 / 4752252 / -0.3% 14042313 / 14164558 / -0.9% 0.34 / 0.34 / +0.6%
tpch_q07/duckdb:vortex-file-compressed 24031053 / 23905696 / +0.5% 39707574 / 40164989 / -1.1% 0.61 / 0.60 / +1.7%
tpch_q08/duckdb:vortex-file-compressed 21025946 / 21394339 / -1.7% 36891018 / 37038669 / -0.4% 0.57 / 0.58 / -1.3%
tpch_q09/duckdb:vortex-file-compressed 21498050 / 21786174 / -1.3% 38234786 / 38276854 / -0.1% 0.56 / 0.57 / -1.2%
tpch_q10/duckdb:vortex-file-compressed 26610738 / 27012064 / -1.5% 41204439 / 41923289 / -1.7% 0.65 / 0.64 / +0.2%
tpch_q11/duckdb:vortex-file-compressed 8307647 / 8303644 / +0.0% 16915138 / 17868992 / -5.3% 0.49 / 0.46 / +5.7%
tpch_q12/duckdb:vortex-file-compressed 14339258 / 14117532 / +1.6% 26668113 / 27344593 / -2.5% 0.54 / 0.52 / +4.1%
tpch_q13/duckdb:vortex-file-compressed 23442141 / 23344085 / +0.4% 32252084 / 32436094 / -0.6% 0.73 / 0.72 / +1.0%
tpch_q14/duckdb:vortex-file-compressed 11477112 / 11537964 / -0.5% 22954323 / 23586045 / -2.7% 0.50 / 0.49 / +2.2%
tpch_q15/duckdb:vortex-file-compressed 8728480 / 8412218 / +3.8% 20357014 / 19967956 / +1.9% 0.43 / 0.42 / +1.8%
tpch_q16/duckdb:vortex-file-compressed 16506751 / 16450203 / +0.3% 23620768 / 24054392 / -1.8% 0.70 / 0.68 / +2.2%
tpch_q17/duckdb:vortex-file-compressed 12344449 / 11939041 / +3.4% 24521372 / 25996187 / -5.7% 0.50 / 0.46 / +9.6%
tpch_q18/duckdb:vortex-file-compressed 23641067 / 23379504 / +1.1% 34868467 / 36488433 / -4.4% 0.68 / 0.64 / +5.8%
tpch_q19/duckdb:vortex-file-compressed 11065052 / 10895028 / +1.6% 22829040 / 23431847 / -2.6% 0.48 / 0.46 / +4.2%
tpch_q20/duckdb:vortex-file-compressed 14279547 / 14047619 / +1.7% 28999455 / 29620105 / -2.1% 0.49 / 0.47 / +3.8%
tpch_q21/duckdb:vortex-file-compressed 36330504 / 36878967 / -1.5% 54598676 / 54556815 / +0.1% 0.67 / 0.68 / -1.6%
tpch_q22/duckdb:vortex-file-compressed 19207882 / 19130524 / +0.4% 28910684 / 29994638 / -3.6% 0.66 / 0.64 / +4.2%
duckdb / parquet / ns (1.002x ➖, 0↑ 0↓)
name hot ns (PR / base / %diff) cold ns (PR / base / %diff) hot/cold (PR / base / %diff)
tpch_q01/duckdb:parquet 48223415 / 48191810 / +0.1% 56329192 / 55076271 / +2.3% 0.86 / 0.88 / -2.2%
tpch_q02/duckdb:parquet 48627222 / 48961127 / -0.7% 55494474 / 56789869 / -2.3% 0.88 / 0.86 / +1.6%
tpch_q03/duckdb:parquet 60655877 / 60629292 / +0.0% 70359636 / 68687216 / +2.4% 0.86 / 0.88 / -2.3%
tpch_q04/duckdb:parquet 35110520 / 35090286 / +0.1% 44248468 / 44044624 / +0.5% 0.79 / 0.80 / -0.4%
tpch_q05/duckdb:parquet 67667373 / 67328631 / +0.5% 78892272 / 77458134 / +1.9% 0.86 / 0.87 / -1.3%
tpch_q06/duckdb:parquet 22059616 / 22239940 / -0.8% 28526514 / 27511945 / +3.7% 0.77 / 0.81 / -4.3%
tpch_q07/duckdb:parquet 64639671 / 64389844 / +0.4% 76173222 / 75397701 / +1.0% 0.85 / 0.85 / -0.6%
tpch_q08/duckdb:parquet 76160042 / 75884068 / +0.4% 88630820 / 86956392 / +1.9% 0.86 / 0.87 / -1.5%
tpch_q09/duckdb:parquet 108433992 / 108361260 / +0.1% 120703833 / 120916486 / -0.2% 0.90 / 0.90 / +0.2%
tpch_q10/duckdb:parquet 106770385 / 106267182 / +0.5% 118212139 / 119574769 / -1.1% 0.90 / 0.89 / +1.6%
tpch_q11/duckdb:parquet 30095964 / 29621509 / +1.6% 35565893 / 35838943 / -0.8% 0.85 / 0.83 / +2.4%
tpch_q12/duckdb:parquet 38097218 / 37986253 / +0.3% 45593037 / 45613822 / -0.0% 0.84 / 0.83 / +0.3%
tpch_q13/duckdb:parquet 175750681 / 175229492 / +0.3% 184287816 / 186674899 / -1.3% 0.95 / 0.94 / +1.6%
tpch_q14/duckdb:parquet 42916234 / 42750953 / +0.4% 51919421 / 52440777 / -1.0% 0.83 / 0.82 / +1.4%
tpch_q15/duckdb:parquet 32981089 / 32981059 / +0.0% 42997583 / 41493065 / +3.6% 0.77 / 0.79 / -3.5%
tpch_q16/duckdb:parquet 47603535 / 47964949 / -0.8% 54004382 / 53450134 / +1.0% 0.88 / 0.90 / -1.8%
tpch_q17/duckdb:parquet 48305533 / 48181865 / +0.3% 58463780 / 59133268 / -1.1% 0.83 / 0.81 / +1.4%
tpch_q18/duckdb:parquet 74745481 / 74734827 / +0.0% 82110902 / 81597852 / +0.6% 0.91 / 0.92 / -0.6%
tpch_q19/duckdb:parquet 48686478 / 48977744 / -0.6% 54849806 / 58850468 / -6.8% 0.89 / 0.83 / +6.7%
tpch_q20/duckdb:parquet 63728607 / 63872186 / -0.2% 74332964 / 74967132 / -0.8% 0.86 / 0.85 / +0.6%
tpch_q21/duckdb:parquet 109136855 / 107603220 / +1.4% 117319721 / 117278383 / +0.0% 0.93 / 0.92 / +1.4%
tpch_q22/duckdb:parquet 50972534 / 50856992 / +0.2% 59615579 / 60488274 / -1.4% 0.86 / 0.84 / +1.7%

No file size changes detected.

@github-actions

github-actions Bot commented Oct 29, 2026 •

Copy link
Copy Markdown
Contributor

Benchmarks: PolarSignals Profiling 📖

Commits: PR 92221095 vs base 2a9c7628
Vortex (geomean): hot 0.999x ➖ · cold 1.194x ❌


datafusion / vortex-file-compressed / ns (0.999x ➖, 0↑ 0↓)
name hot ns (PR / base / %diff) cold ns (PR / base / %diff) hot/cold (PR / base / %diff)
polarsignals_q00/datafusion:vortex-file-compressed 44503745 / 44621484 / -0.3% 96802911 / 85053639 / +13.8% 🔴 0.46 / 0.52 / -12.4%
polarsignals_q01/datafusion:vortex-file-compressed 50462532 / 50679540 / -0.4% 120169869 / 106093794 / +13.3% 🔴 0.42 / 0.48 / -12.1%
polarsignals_q02/datafusion:vortex-file-compressed 43732425 / 43581214 / +0.3% 94697353 / 81980896 / +15.5% 🔴 0.46 / 0.53 / -13.1%
polarsignals_q03/datafusion:vortex-file-compressed 53338805 / 53919581 / -1.1% 124859856 / 107832012 / +15.8% 🔴 0.43 / 0.50 / -14.6%
polarsignals_q04/datafusion:vortex-file-compressed 9101254 / 9258392 / -1.7% 52812935 / 42077120 / +25.5% 🔴 0.17 / 0.22 / -21.7%
polarsignals_q05/datafusion:vortex-file-compressed 11780445 / 11675713 / +0.9% 57756273 / 45170160 / +27.9% 🔴 0.20 / 0.26 / -21.1%
polarsignals_q06/datafusion:vortex-file-compressed 23325119 / 23641958 / -1.3% 67342870 / 54123616 / +24.4% 🔴 0.35 / 0.44 / -20.7%
polarsignals_q07/datafusion:vortex-file-compressed 11205417 / 10929114 / +2.5% 54484254 / 44518959 / +22.4% 🔴 0.21 / 0.25 / -16.2%
polarsignals_q08/datafusion:vortex-file-compressed 98463974 / 97961134 / +0.5% 163378152 / 152512854 / +7.1% 0.60 / 0.64 / -6.2%
polarsignals_q09/datafusion:vortex-file-compressed 10099096 / 10175232 / -0.7% 52004124 / 39921233 / +30.3% 🔴 0.19 / 0.25 / -23.8%

No file size changes detected.

@github-actions

github-actions Bot commented Oct 29, 2026 •

Copy link
Copy Markdown
Contributor

🚨🚨🚨❌❌❌ SQL BENCHMARK FAILED ❌❌❌🚨🚨🚨

Benchmark TPC-DS SF=1 on NVME (pr) failed! Check the workflow run for details.

@github-actions

github-actions Bot commented Oct 29, 2026 •

Copy link
Copy Markdown
Contributor

🚨🚨🚨❌❌❌ SQL BENCHMARK FAILED ❌❌❌🚨🚨🚨

Benchmark TPC-H SF=10 on NVME (pr) failed! Check the workflow run for details.

@github-actions

github-actions Bot commented Oct 29, 2026 •

Copy link
Copy Markdown
Contributor

🚨🚨🚨❌❌❌ SQL BENCHMARK FAILED ❌❌❌🚨🚨🚨

Benchmark Clickbench on NVME (pr) failed! Check the workflow run for details.

@github-actions

github-actions Bot commented Oct 29, 2026 •

Copy link
Copy Markdown
Contributor

Benchmarks: TPC-H SF=1 on S3 📖

Commits: PR 92221095 vs base 1d09649b
Verdict: No clear signal (environment too noisy confidence)
Attributed Vortex impact: +5.6%
Engines: DataFusion No clear signal (+3.5%, environment too noisy confidence) · DuckDB No clear signal (+7.6%, environment too noisy confidence)
Vortex (geomean): hot 1.064x ➖ · cold 1.101x ➖
Parquet (geomean): hot 1.008x ➖ · cold 1.009x ➖
Shifts: Parquet (control) +0.8% · Median polish +3.5%

How to read Verdict and Engines
  • Verdict: Overall PR-level signal after subtracting baseline drift estimated from Parquet control rows. It can be Likely improvement, Likely regression, or No clear signal.
  • Engines: Per-engine attribution. DataFusion is compared against DataFusion/Parquet controls; DuckDB is compared against DuckDB/Parquet controls. This answers whether each engine improved or regressed independently.
  • Confidence: Based on directional consistency, share of rows above the noise floor, and control-run noise.
  • Hot vs cold: Every measurement is run several times. The first run is reported as the cold run, and the median of the runs after it is reported as the hot run. The verdict and significance use hot runs; each target's geomean reports hot and cold beside each other where the individual runs were recorded and hot alone where they were not, the cold column shows first-run cost per row, and hot/cold is how much of each run the warm path saves. Rows whose results predate per-run reporting show only one value, taken from the value the runner reported.
  • Table cells: Each cell reads PR / base / %diff. The hot and cold columns mark a change as 🔴 slower or 🟢 faster once it clears this suite's threshold; anything smaller is noise here and is left unmarked, as is hot/cold, because a shift in the warm-up ratio is not a win or a loss by itself.

datafusion / vortex-file-compressed / ns (1.024x ➖, 1↑ 3↓)
name hot ns (PR / base / %diff) cold ns (PR / base / %diff) hot/cold (PR / base / %diff)
tpch_q01/datafusion:vortex-file-compressed 350530345 / 222154254 / +57.8% 🔴 1212043832 / 814591975 / +48.8% 🔴 0.29 / 0.27 / +6.0%
tpch_q02/datafusion:vortex-file-compressed 350959859 / 315432783 / +11.3% 675072235 / 727695886 / -7.2% 0.52 / 0.43 / +19.9%
tpch_q03/datafusion:vortex-file-compressed 354166036 / 399892039 / -11.4% 982754101 / 1181390062 / -16.8% 0.36 / 0.34 / +6.5%
tpch_q04/datafusion:vortex-file-compressed 223355612 / 186319978 / +19.9% 444974187 / 429069645 / +3.7% 0.50 / 0.43 / +15.6%
tpch_q05/datafusion:vortex-file-compressed 430229045 / 472592334 / -9.0% 663551973 / 664010317 / -0.1% 0.65 / 0.71 / -8.9%
tpch_q06/datafusion:vortex-file-compressed 146657809 / 176515226 / -16.9% 306193240 / 281077299 / +8.9% 0.48 / 0.63 / -23.7%
tpch_q07/datafusion:vortex-file-compressed 406344553 / 494983926 / -17.9% 606928231 / 683240062 / -11.2% 0.67 / 0.72 / -7.6%
tpch_q08/datafusion:vortex-file-compressed 482006066 / 569435616 / -15.4% 727791370 / 618500176 / +17.7% 0.66 / 0.92 / -28.1%
tpch_q09/datafusion:vortex-file-compressed 324455680 / 312222108 / +3.9% 579209693 / 500763591 / +15.7% 0.56 / 0.62 / -10.2%
tpch_q10/datafusion:vortex-file-compressed 373063268 / 368560620 / +1.2% 496733012 / 572586696 / -13.2% 0.75 / 0.64 / +16.7%
tpch_q11/datafusion:vortex-file-compressed 300361773 / 330047066 / -9.0% 474096883 / 540174172 / -12.2% 0.63 / 0.61 / +3.7%
tpch_q12/datafusion:vortex-file-compressed 209972823 / 216573816 / -3.0% 384616318 / 470388516 / -18.2% 0.55 / 0.46 / +18.6%
tpch_q13/datafusion:vortex-file-compressed 142459703 / 136971641 / +4.0% 423138014 / 362379625 / +16.8% 0.34 / 0.38 / -10.9%
tpch_q14/datafusion:vortex-file-compressed 183872852 / 195691008 / -6.0% 358810190 / 321568265 / +11.6% 0.51 / 0.61 / -15.8%
tpch_q15/datafusion:vortex-file-compressed 583543715 / 356332959 / +63.8% 🔴 673015828 / 571187556 / +17.8% 0.87 / 0.62 / +39.0%
tpch_q16/datafusion:vortex-file-compressed 147702192 / 128887070 / +14.6% 339695381 / 270501395 / +25.6% 0.43 / 0.48 / -8.7%
tpch_q17/datafusion:vortex-file-compressed 312371999 / 248322567 / +25.8% 519439959 / 458068313 / +13.4% 0.60 / 0.54 / +10.9%
tpch_q18/datafusion:vortex-file-compressed 297582186 / 321125311 / -7.3% 723213149 / 428850585 / +68.6% 🔴 0.41 / 0.75 / -45.0%
tpch_q19/datafusion:vortex-file-compressed 323542166 / 238307664 / +35.8% 🔴 357936100 / 381381483 / -6.1% 0.90 / 0.62 / +44.7%
tpch_q20/datafusion:vortex-file-compressed 340713649 / 317047949 / +7.5% 501256454 / 664297461 / -24.5% 0.68 / 0.48 / +42.4%
tpch_q21/datafusion:vortex-file-compressed 497233006 / 858135517 / -42.1% 🟢 664348901 / 830494854 / -20.0% 0.75 / 1.03 / -27.6%
tpch_q22/datafusion:vortex-file-compressed 185511064 / 180910474 / +2.5% 500979945 / 402785646 / +24.4% 0.37 / 0.45 / -17.6%
datafusion / parquet / ns (0.989x ➖, 0↑ 2↓)
name hot ns (PR / base / %diff) cold ns (PR / base / %diff) hot/cold (PR / base / %diff)
tpch_q01/datafusion:parquet 158424219 / 161593667 / -2.0% 544145019 / 486618857 / +11.8% 0.29 / 0.33 / -12.3%
tpch_q02/datafusion:parquet 316584769 / 296179085 / +6.9% 568031969 / 511960868 / +11.0% 0.56 / 0.58 / -3.7%
tpch_q03/datafusion:parquet 254556774 / 312603552 / -18.6% 773658160 / 723847616 / +6.9% 0.33 / 0.43 / -23.8%
tpch_q04/datafusion:parquet 164352110 / 167910435 / -2.1% 322772761 / 361042949 / -10.6% 0.51 / 0.47 / +9.5%
tpch_q05/datafusion:parquet 357312950 / 372233340 / -4.0% 475875066 / 594650028 / -20.0% 0.75 / 0.63 / +20.0%
tpch_q06/datafusion:parquet 118151533 / 125384522 / -5.8% 462650604 / 257572903 / +79.6% 🔴 0.26 / 0.49 / -47.5%
tpch_q07/datafusion:parquet 351600085 / 342764669 / +2.6% 450094975 / 577174571 / -22.0% 0.78 / 0.59 / +31.5%
tpch_q08/datafusion:parquet 768780616 / 417579974 / +84.1% 🔴 738161330 / 514316467 / +43.5% 🔴 1.04 / 0.81 / +28.3%
tpch_q09/datafusion:parquet 403836641 / 488127593 / -17.3% 531810637 / 634913752 / -16.2% 0.76 / 0.77 / -1.2%
tpch_q10/datafusion:parquet 304275208 / 326793997 / -6.9% 466782733 / 441088464 / +5.8% 0.65 / 0.74 / -12.0%
tpch_q11/datafusion:parquet 259648290 / 254644930 / +2.0% 358798985 / 428318375 / -16.2% 0.72 / 0.59 / +21.7%
tpch_q12/datafusion:parquet 156595613 / 198515620 / -21.1% 308444912 / 316212167 / -2.5% 0.51 / 0.63 / -19.1%
tpch_q13/datafusion:parquet 280573619 / 278152836 / +0.9% 430544175 / 730173776 / -41.0% 🟢 0.65 / 0.38 / +71.1%
tpch_q14/datafusion:parquet 177498776 / 252455217 / -29.7% 345769933 / 284199798 / +21.7% 0.51 / 0.89 / -42.2%
tpch_q15/datafusion:parquet 324481761 / 351433287 / -7.7% 480863545 / 481531458 / -0.1% 0.67 / 0.73 / -7.5%
tpch_q16/datafusion:parquet 104006442 / 131273094 / -20.8% 236670525 / 278832615 / -15.1% 0.44 / 0.47 / -6.7%
tpch_q17/datafusion:parquet 292977772 / 318523735 / -8.0% 470598193 / 482463179 / -2.5% 0.62 / 0.66 / -5.7%
tpch_q18/datafusion:parquet 400513998 / 326277460 / +22.8% 507646060 / 457527535 / +11.0% 0.79 / 0.71 / +10.6%
tpch_q19/datafusion:parquet 180441212 / 181023427 / -0.3% 271389859 / 302568354 / -10.3% 0.66 / 0.60 / +11.1%
tpch_q20/datafusion:parquet 251792209 / 255571406 / -1.5% 361813362 / 358613141 / +0.9% 0.70 / 0.71 / -2.4%
tpch_q21/datafusion:parquet 553522733 / 314736185 / +75.9% 🔴 456236985 / 437118870 / +4.4% 1.21 / 0.72 / +68.5%
tpch_q22/datafusion:parquet 151188817 / 172622826 / -12.4% 337988851 / 319903113 / +5.7% 0.45 / 0.54 / -17.1%
duckdb / vortex-file-compressed / ns (1.105x ➖, 0↑ 3↓)
name hot ns (PR / base / %diff) cold ns (PR / base / %diff) hot/cold (PR / base / %diff)
tpch_q01/duckdb:vortex-file-compressed 254563427 / 240658133 / +5.8% 471487050 / 306491136 / +53.8% 🔴 0.54 / 0.79 / -31.2%
tpch_q02/duckdb:vortex-file-compressed 434585004 / 400621849 / +8.5% 467264897 / 465465734 / +0.4% 0.93 / 0.86 / +8.1%
tpch_q03/duckdb:vortex-file-compressed 427714085 / 402146077 / +6.4% 446976798 / 460543640 / -2.9% 0.96 / 0.87 / +9.6%
tpch_q04/duckdb:vortex-file-compressed 249248297 / 245443218 / +1.6% 296740885 / 301022196 / -1.4% 0.84 / 0.82 / +3.0%
tpch_q05/duckdb:vortex-file-compressed 591223243 / 518820424 / +14.0% 615216219 / 569655285 / +8.0% 0.96 / 0.91 / +5.5%
tpch_q06/duckdb:vortex-file-compressed 175655120 / 162123437 / +8.3% 266104178 / 206523017 / +28.8% 0.66 / 0.79 / -15.9%
tpch_q07/duckdb:vortex-file-compressed 425831460 / 397090485 / +7.2% 506744173 / 391003889 / +29.6% 0.84 / 1.02 / -17.3%
tpch_q08/duckdb:vortex-file-compressed 753406522 / 549467255 / +37.1% 🔴 818364440 / 647180890 / +26.5% 0.92 / 0.85 / +8.4%
tpch_q09/duckdb:vortex-file-compressed 440738062 / 409980961 / +7.5% 513755115 / 473296413 / +8.5% 0.86 / 0.87 / -1.0%
tpch_q10/duckdb:vortex-file-compressed 711375462 / 506597721 / +40.4% 🔴 707635252 / 506502726 / +39.7% 🔴 1.01 / 1.00 / +0.5%
tpch_q11/duckdb:vortex-file-compressed 290330989 / 275836094 / +5.3% 547884930 / 326266330 / +67.9% 🔴 0.53 / 0.85 / -37.3%
tpch_q12/duckdb:vortex-file-compressed 265357123 / 249303256 / +6.4% 308646619 / 435704808 / -29.2% 0.86 / 0.57 / +50.3%
tpch_q13/duckdb:vortex-file-compressed 276429382 / 267826106 / +3.2% 241590634 / 424412220 / -43.1% 🟢 1.14 / 0.63 / +81.3%
tpch_q14/duckdb:vortex-file-compressed 236852733 / 229555717 / +3.2% 291305060 / 312807618 / -6.9% 0.81 / 0.73 / +10.8%
tpch_q15/duckdb:vortex-file-compressed 253342521 / 233138227 / +8.7% 294936305 / 267945943 / +10.1% 0.86 / 0.87 / -1.3%
tpch_q16/duckdb:vortex-file-compressed 213490978 / 193523278 / +10.3% 279136375 / 235506661 / +18.5% 0.76 / 0.82 / -6.9%
tpch_q17/duckdb:vortex-file-compressed 492453819 / 296774302 / +65.9% 🔴 509733209 / 339106123 / +50.3% 🔴 0.97 / 0.88 / +10.4%
tpch_q18/duckdb:vortex-file-compressed 433279000 / 355309282 / +21.9% 593202055 / 432636316 / +37.1% 🔴 0.73 / 0.82 / -11.1%
tpch_q19/duckdb:vortex-file-compressed 379532737 / 299108202 / +26.9% 530547903 / 386812262 / +37.2% 🔴 0.72 / 0.77 / -7.5%
tpch_q20/duckdb:vortex-file-compressed 408495296 / 452792082 / -9.8% 448819387 / 398350349 / +12.7% 0.91 / 1.14 / -19.9%
tpch_q21/duckdb:vortex-file-compressed 758738571 / 842892293 / -10.0% 887162738 / 595110658 / +49.1% 🔴 0.86 / 1.42 / -39.6%
tpch_q22/duckdb:vortex-file-compressed 299825218 / 337713131 / -11.2% 426037349 / 314862244 / +35.3% 🔴 0.70 / 1.07 / -34.4%
duckdb / parquet / ns (1.027x ➖, 0↑ 0↓)
name hot ns (PR / base / %diff) cold ns (PR / base / %diff) hot/cold (PR / base / %diff)
tpch_q01/duckdb:parquet 201679712 / 227376317 / -11.3% 334873803 / 264289037 / +26.7% 0.60 / 0.86 / -30.0%
tpch_q02/duckdb:parquet 455880542 / 452369880 / +0.8% 470902689 / 565837534 / -16.8% 0.97 / 0.80 / +21.1%
tpch_q03/duckdb:parquet 444421002 / 436893759 / +1.7% 456157170 / 476090676 / -4.2% 0.97 / 0.92 / +6.2%
tpch_q04/duckdb:parquet 315046766 / 248899693 / +26.6% 324884624 / 369390205 / -12.0% 0.97 / 0.67 / +43.9%
tpch_q05/duckdb:parquet 739860792 / 576683354 / +28.3% 867282846 / 624300088 / +38.9% 🔴 0.85 / 0.92 / -7.6%
tpch_q06/duckdb:parquet 137363538 / 169982086 / -19.2% 179989105 / 225398279 / -20.1% 0.76 / 0.75 / +1.2%
tpch_q07/duckdb:parquet 539886165 / 530631594 / +1.7% 581928983 / 492800368 / +18.1% 0.93 / 1.08 / -13.8%
tpch_q08/duckdb:parquet 692159242 / 675359248 / +2.5% 707518976 / 725877378 / -2.5% 0.98 / 0.93 / +5.1%
tpch_q09/duckdb:parquet 583121589 / 578342843 / +0.8% 569162009 / 665323694 / -14.5% 1.02 / 0.87 / +17.9%
tpch_q10/duckdb:parquet 709084412 / 670886777 / +5.7% 678816635 / 1121448603 / -39.5% 🟢 1.04 / 0.60 / +74.6%
tpch_q11/duckdb:parquet 291864857 / 302321127 / -3.5% 407516074 / 332051851 / +22.7% 0.72 / 0.91 / -21.3%
tpch_q12/duckdb:parquet 292956754 / 301458001 / -2.8% 320449043 / 406833087 / -21.2% 0.91 / 0.74 / +23.4%
tpch_q13/duckdb:parquet 380423216 / 407069876 / -6.5% 484198238 / 545052499 / -11.2% 0.79 / 0.75 / +5.2%
tpch_q14/duckdb:parquet 340618404 / 276619645 / +23.1% 405850321 / 426162473 / -4.8% 0.84 / 0.65 / +29.3%
tpch_q15/duckdb:parquet 196508577 / 231841083 / -15.2% 238295766 / 324101916 / -26.5% 0.82 / 0.72 / +15.3%
tpch_q16/duckdb:parquet 285113548 / 267123368 / +6.7% 334011074 / 277727422 / +20.3% 0.85 / 0.96 / -11.3%
tpch_q17/duckdb:parquet 278559001 / 331754906 / -16.0% 360502474 / 389396937 / -7.4% 0.77 / 0.85 / -9.3%
tpch_q18/duckdb:parquet 365178264 / 334871830 / +9.1% 471424124 / 369943987 / +27.4% 0.77 / 0.91 / -14.4%
tpch_q19/duckdb:parquet 324448343 / 275278774 / +17.9% 361153847 / 301595298 / +19.7% 0.90 / 0.91 / -1.6%
tpch_q20/duckdb:parquet 475289430 / 469341125 / +1.3% 521087018 / 451727350 / +15.4% 0.91 / 1.04 / -12.2%
tpch_q21/duckdb:parquet 532038436 / 492317819 / +8.1% 806825038 / 487389597 / +65.5% 🔴 0.66 / 1.01 / -34.7%
tpch_q22/duckdb:parquet 326777568 / 280238067 / +16.6% 419787006 / 290782193 / +44.4% 🔴 0.78 / 0.96 / -19.2%

@github-actions

github-actions Bot commented Oct 29, 2026 •

Copy link
Copy Markdown
Contributor

Benchmarks: FineWeb NVMe 📖

Commits: PR 92221095 vs base 2a9c7628
Verdict: No clear signal (low confidence)
Attributed Vortex impact: +1.5%
Engines: DataFusion No clear signal (+2.1%, low confidence) · DuckDB No clear signal (+1.0%, low confidence)
Vortex (geomean): hot 1.012x ➖ · cold 1.070x ➖
Parquet (geomean): hot 0.997x ➖ · cold 1.056x ➖
Shifts: Parquet (control) -0.3% · Median polish +0.4%

How to read Verdict and Engines
  • Verdict: Overall PR-level signal after subtracting baseline drift estimated from Parquet control rows. It can be Likely improvement, Likely regression, or No clear signal.
  • Engines: Per-engine attribution. DataFusion is compared against DataFusion/Parquet controls; DuckDB is compared against DuckDB/Parquet controls. This answers whether each engine improved or regressed independently.
  • Confidence: Based on directional consistency, share of rows above the noise floor, and control-run noise.
  • Hot vs cold: Every measurement is run several times. The first run is reported as the cold run, and the median of the runs after it is reported as the hot run. The verdict and significance use hot runs; each target's geomean reports hot and cold beside each other where the individual runs were recorded and hot alone where they were not, the cold column shows first-run cost per row, and hot/cold is how much of each run the warm path saves. Rows whose results predate per-run reporting show only one value, taken from the value the runner reported.
  • Table cells: Each cell reads PR / base / %diff. The hot and cold columns mark a change as 🔴 slower or 🟢 faster once it clears this suite's threshold; anything smaller is noise here and is left unmarked, as is hot/cold, because a shift in the warm-up ratio is not a win or a loss by itself.

datafusion / vortex-file-compressed / ns (1.019x ➖, 0↑ 0↓)
name hot ns (PR / base / %diff) cold ns (PR / base / %diff) hot/cold (PR / base / %diff)
fineweb_q00/datafusion:vortex-file-compressed 7022090 / 6903610 / +1.7% 18525930 / 17910056 / +3.4% 0.38 / 0.39 / -1.7%
fineweb_q01/datafusion:vortex-file-compressed 8667609 / 8689116 / -0.2% 50760152 / 44097276 / +15.1% 🔴 0.17 / 0.20 / -13.3%
fineweb_q02/datafusion:vortex-file-compressed 10649754 / 10214480 / +4.3% 52516995 / 51600306 / +1.8% 0.20 / 0.20 / +2.4%
fineweb_q03/datafusion:vortex-file-compressed 19969944 / 19591272 / +1.9% 117777751 / 107891043 / +9.2% 0.17 / 0.18 / -6.6%
fineweb_q04/datafusion:vortex-file-compressed 66817353 / 66685910 / +0.2% 159598452 / 160241075 / -0.4% 0.42 / 0.42 / +0.6%
fineweb_q05/datafusion:vortex-file-compressed 69751919 / 67421159 / +3.5% 158697283 / 157717787 / +0.6% 0.44 / 0.43 / +2.8%
fineweb_q06/datafusion:vortex-file-compressed 19761698 / 18932958 / +4.4% 127615203 / 114653888 / +11.3% 🔴 0.15 / 0.17 / -6.2%
fineweb_q07/datafusion:vortex-file-compressed 17823652 / 18335188 / -2.8% 120523460 / 105325486 / +14.4% 🔴 0.15 / 0.17 / -15.0%
fineweb_q08/datafusion:vortex-file-compressed 10656533 / 10204749 / +4.4% 49037234 / 50634223 / -3.2% 0.22 / 0.20 / +7.8%
datafusion / parquet / ns (0.998x ➖, 0↑ 0↓)
name hot ns (PR / base / %diff) cold ns (PR / base / %diff) hot/cold (PR / base / %diff)
fineweb_q00/datafusion:parquet 9238616 / 9482157 / -2.6% 33770627 / 32111399 / +5.2% 0.27 / 0.30 / -7.4%
fineweb_q01/datafusion:parquet 75843778 / 75129645 / +1.0% 174332622 / 156360166 / +11.5% 🔴 0.44 / 0.48 / -9.5%
fineweb_q02/datafusion:parquet 74955700 / 73747780 / +1.6% 172298443 / 159846521 / +7.8% 0.44 / 0.46 / -5.7%
fineweb_q03/datafusion:parquet 74081856 / 74768342 / -0.9% 167443707 / 157975411 / +6.0% 0.44 / 0.47 / -6.5%
fineweb_q04/datafusion:parquet 82363323 / 82819587 / -0.6% 174523645 / 157992861 / +10.5% 🔴 0.47 / 0.52 / -10.0%
fineweb_q05/datafusion:parquet 80506031 / 81080730 / -0.7% 170221481 / 157515065 / +8.1% 0.47 / 0.51 / -8.1%
fineweb_q06/datafusion:parquet 75602532 / 76820271 / -1.6% 174525694 / 167173020 / +4.4% 0.43 / 0.46 / -5.7%
fineweb_q07/datafusion:parquet 75093128 / 73624110 / +2.0% 169338704 / 156859062 / +8.0% 0.44 / 0.47 / -5.5%
fineweb_q08/datafusion:parquet 73650260 / 73579522 / +0.1% 178748847 / 161416701 / +10.7% 🔴 0.41 / 0.46 / -9.6%
duckdb / vortex-file-compressed / ns (1.006x ➖, 0↑ 0↓)
name hot ns (PR / base / %diff) cold ns (PR / base / %diff) hot/cold (PR / base / %diff)
fineweb_q00/duckdb:vortex-file-compressed 2089345 / 2060458 / +1.4% 7182699 / 7235617 / -0.7% 0.29 / 0.28 / +2.1%
fineweb_q01/duckdb:vortex-file-compressed 9618356 / 9533635 / +0.9% 38710406 / 33428263 / +15.8% 🔴 0.25 / 0.29 / -12.9%
fineweb_q02/duckdb:vortex-file-compressed 10812825 / 10827112 / -0.1% 35631799 / 33795730 / +5.4% 0.30 / 0.32 / -5.3%
fineweb_q03/duckdb:vortex-file-compressed 19905722 / 19616776 / +1.5% 101685611 / 88574457 / +14.8% 🔴 0.20 / 0.22 / -11.6%
fineweb_q04/duckdb:vortex-file-compressed 68931256 / 66064766 / +4.3% 118136841 / 108051698 / +9.3% 0.58 / 0.61 / -4.6%
fineweb_q05/duckdb:vortex-file-compressed 65798082 / 66540793 / -1.1% 117307374 / 105711829 / +11.0% 🔴 0.56 / 0.63 / -10.9%
fineweb_q06/duckdb:vortex-file-compressed 19416564 / 19522226 / -0.5% 104705603 / 94334392 / +11.0% 🔴 0.19 / 0.21 / -10.4%
fineweb_q07/duckdb:vortex-file-compressed 18764431 / 18991482 / -1.2% 100336589 / 88019344 / +14.0% 🔴 0.19 / 0.22 / -13.3%
fineweb_q08/duckdb:vortex-file-compressed 5505798 / 5492126 / +0.2% 14319779 / 14753686 / -2.9% 0.38 / 0.37 / +3.3%
duckdb / parquet / ns (0.996x ➖, 0↑ 0↓)
name hot ns (PR / base / %diff) cold ns (PR / base / %diff) hot/cold (PR / base / %diff)
fineweb_q00/duckdb:parquet 23884379 / 23448581 / +1.9% 28611251 / 27593808 / +3.7% 0.83 / 0.85 / -1.8%
fineweb_q01/duckdb:parquet 55368082 / 54933198 / +0.8% 79996570 / 76123605 / +5.1% 0.69 / 0.72 / -4.1%
fineweb_q02/duckdb:parquet 58168528 / 58051781 / +0.2% 83373517 / 79202689 / +5.3% 0.70 / 0.73 / -4.8%
fineweb_q03/duckdb:parquet 118005634 / 118155770 / -0.1% 187005710 / 179057039 / +4.4% 0.63 / 0.66 / -4.4%
fineweb_q04/duckdb:parquet 159977274 / 161386539 / -0.9% 220592383 / 216551920 / +1.9% 0.73 / 0.75 / -2.7%
fineweb_q05/duckdb:parquet 155821570 / 157119474 / -0.8% 204505042 / 208784938 / -2.0% 0.76 / 0.75 / +1.2%
fineweb_q06/duckdb:parquet 85896287 / 89287572 / -3.8% 156965307 / 154201019 / +1.8% 0.55 / 0.58 / -5.5%
fineweb_q07/duckdb:parquet 89119348 / 91067884 / -2.1% 155463279 / 145204462 / +7.1% 0.57 / 0.63 / -8.6%
fineweb_q08/duckdb:parquet 29658220 / 29225694 / +1.5% 30967138 / 30385447 / +1.9% 0.96 / 0.96 / -0.4%

No file size changes detected.

@codspeed

codspeed Bot commented Oct 29, 2026 •

Copy link
Copy Markdown

Merging this PR will degrade performance by 27.05%

2 warnings

⚠️ Walltime results may be unreliable

Some benchmarks used the Walltime instrument outside CodSpeed Macro Runners, which can make the results inconsistent.

Learn more about Macro Runners

See affected benchmarks

⚠️ 1 benchmark is dominated by system calls

System calls cannot be simulated consistently and are left out of the measure. The reported value understates the real cost. Switch to the Walltime instrument to measure them.

Learn more about system calls

See affected benchmarks

❌ 4 regressed benchmarks
✅ 2300 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

Mode Benchmark BASE HEAD Efficiency
❌ Walltime threshold_neon[64] 2.6 µs 4.2 µs -38.48%
❌ Walltime threshold_neon[80] 3.5 µs 4.7 µs -25.59%
❌ Walltime filter_neon[64] 1.9 µs 2.5 µs -24.7%
❌ Walltime filter_neon[80] 2.6 µs 3.1 µs -17.85%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.


Comparing ji/codex/clickbench-pruned-parallelism (9222109) with develop (e4dce8b)

Open in CodSpeed

Signed-off-by: Joe Isaacs <joe.isaacs@live.co.uk>
Signed-off-by: Joe Isaacs <joe.isaacs@live.co.uk>
@joseph-isaacs joseph-isaacs changed the title Keep heavily pruned V2 scans parallel Tune V2 scan tasks for sparse filters and unfiltered projections Oct 29, 2026
@joseph-isaacs
joseph-isaacs force-pushed the ji/codex/clickbench-pruned-parallelism branch from a3d0ea3 to 9222109 Compare October 29, 2026 16:16
@joseph-isaacs joseph-isaacs added the action/bench-sql Run SQL benchmarks without Vortex Compact on this PR label Oct 29, 2026 — with ChatGPT Codex Connector
@github-actions github-actions Bot removed the action/bench-sql Run SQL benchmarks without Vortex Compact on this PR label Oct 29, 2026
@joseph-isaacs
joseph-isaacs marked this pull request as ready for review October 29, 2026 16:24
joseph-isaacs added a commit that referenced this pull request Oct 29, 2026
Reconcile with merged #10799 and leave overlapping scheduling and read
ownership policies to #10789, #10857, and #10969. Preserve current develop's
structural cache, sparse dictionary threshold, chunk sizing, short-edge
merging, retained segments, and benchmark defaults.

Signed-off-by: Joe Isaacs <joe.isaacs@live.co.uk>
@github-actions

Copy link
Copy Markdown
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

@github-actions github-actions Bot added the stale This PR is stale and will be auto-closed soon label Nov 13, 2026

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

changelog/performance A performance improvement stale This PR is stale and will be auto-closed soon

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant