Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

feat: ChunkedArray uses a builder to implement to_canonical #2511

Merged
merged 5 commits into from
Mar 7, 2025
Merged

feat: ChunkedArray uses a builder to implement to_canonical #2511

merged 5 commits into from
Mar 7, 2025

Conversation

danking
Copy link
Member

@danking danking commented Feb 25, 2025

No description provided.

@danking danking marked this pull request as ready for review February 25, 2025 17:28
@danking danking added the benchmark Run benchmarks on this branch label Feb 25, 2025
@danking danking enabled auto-merge (squash) February 25, 2025 17:28
@github-actions github-actions bot removed the benchmark Run benchmarks on this branch label Feb 25, 2025
@danking danking mentioned this pull request Feb 25, 2025
Closed
@codspeed-hq CodSpeed HQ
Copy link

codspeed-hq bot commented Feb 25, 2025 •
edited
Loading

CodSpeed Performance Report

Merging #2511 will degrade performances by 19.58%

Comparing dk/chunked-array-into-canonical-canonical-into-2 (41d62cc) with develop (8ba5f27)

Summary

⚡ 42 improvements
❌ 4 regressions
✅ 729 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

Benchmark BASE HEAD Change
⚡ chunked_dict_fsst_into_canonical[(1000, 1000, 10)] 1.3 ms 1.2 ms +10.05%
⚡ chunked_dict_primitive_into_canonical[f32, (1000, 10, 10)] 186.4 µs 86.7 µs ×2.1
⚡ chunked_dict_primitive_into_canonical[f32, (1000, 10, 100)] 1,593.1 µs 717.2 µs ×2.2
⚡ chunked_dict_primitive_into_canonical[f32, (1000, 100, 10)] 186.1 µs 88.6 µs ×2.1
⚡ chunked_dict_primitive_into_canonical[f32, (1000, 100, 100)] 1,628.5 µs 733.5 µs ×2.2
⚡ chunked_dict_primitive_into_canonical[f32, (1000, 1000, 10)] 205.6 µs 105.1 µs +95.53%
⚡ chunked_dict_primitive_into_canonical[f32, (1000, 1000, 100)] 1,779.2 µs 896.7 µs +98.42%
⚡ chunked_dict_primitive_into_canonical[f64, (1000, 10, 10)] 212.2 µs 105.2 µs ×2
⚡ chunked_dict_primitive_into_canonical[f64, (1000, 10, 100)] 1,849.1 µs 901.3 µs ×2.1
⚡ chunked_dict_primitive_into_canonical[f64, (1000, 100, 10)] 215.9 µs 108.5 µs +98.99%
⚡ chunked_dict_primitive_into_canonical[f64, (1000, 100, 100)] 1,903.9 µs 933.6 µs ×2
⚡ chunked_dict_primitive_into_canonical[f64, (1000, 1000, 10)] 250.2 µs 141.2 µs +77.13%
⚡ chunked_dict_primitive_into_canonical[f64, (1000, 1000, 100)] 2.2 ms 1.3 ms +75.78%
⚡ chunked_dict_primitive_into_canonical[u32, (1000, 10, 10)] 184.1 µs 86.6 µs ×2.1
⚡ chunked_dict_primitive_into_canonical[u32, (1000, 10, 100)] 1,611.1 µs 717.4 µs ×2.2
⚡ chunked_dict_primitive_into_canonical[u32, (1000, 100, 10)] 188.1 µs 88.3 µs ×2.1
⚡ chunked_dict_primitive_into_canonical[u32, (1000, 100, 100)] 1,627.7 µs 733.3 µs ×2.2
⚡ chunked_dict_primitive_into_canonical[u32, (1000, 1000, 10)] 206.2 µs 104.8 µs +96.79%
⚡ chunked_dict_primitive_into_canonical[u32, (1000, 1000, 100)] 1,778.5 µs 896.5 µs +98.39%
⚡ chunked_dict_primitive_into_canonical[u64, (1000, 10, 10)] 217.1 µs 105 µs ×2.1
... ... ... ... ...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

@danking
Copy link
Member Author

danking commented Feb 25, 2025

I think we should merge this. One of the regressions seems flaky. I've seen it before on unrelated code. The other one is probably real. I'm not entirely sure the cause, but the profile is dominated by the cost to filter patches when the Mask is 1-ε full. Our implementation for that is not so great.

@gatesn
Copy link
Contributor

gatesn commented Feb 25, 2025 •
edited
Loading

The big difference here is that into_builder is a recursive canonicalize, whereas to_canonical only canonicalizes a single level.

So before, a Chunked gets cheaply turned into a Struct. Now it does a full recursive decompression.

I think it might be benefital to retain the distinction between the two for now? Not sure though...

@danking
Copy link
Member Author

danking commented Feb 25, 2025

Hmm. The benchmark does not use chunked arrays (in develop or in this PR). I am a bit confused how the changes here could possibly have affected it.

@danking
Copy link
Member Author

danking commented Feb 25, 2025

I think the microbenchmark is a red herring. The profiles look very similar, albeit reliably slower on this PR. ChunkedArray does not appear anywhere in the profile. The only difference I've found is that the left parts dictionary is reordered. I don't know why this would cause a slow down though.

@robert3005
Copy link
Member

robert3005 commented Feb 25, 2025 •
edited
Loading

the benchmarks that improved were using gen_dict_primitive_chunks which collects to ChunkedArray. I think what @gatesn says seems correct. Though there's no Structs in the benchmarks :/ Ignore my comment I focused on benchmarks and not on the semantics

@danking
Copy link
Member Author

danking commented Feb 25, 2025

blocked by #2514

@danking danking force-pushed the dk/chunked-array-into-canonical-canonical-into-2 branch from e8c8cad to 4035433 Compare February 26, 2025 14:37
@danking danking added the benchmark Run benchmarks on this branch label Feb 26, 2025
@github-actions github-actions bot removed the benchmark Run benchmarks on this branch label Feb 26, 2025
@danking danking added benchmark Run benchmarks on this branch benchmark-sql labels Feb 26, 2025
@github-actions github-actions bot removed benchmark Run benchmarks on this branch benchmark-sql labels Feb 26, 2025
@github-actions GitHub Actions
Copy link
Contributor

github-actions bot commented Feb 26, 2025 •
edited
Loading

Benchmarks: random_access

Table of Results
name PR 2e01f59 base 76bc8eb ratio (PR/base) unit
random-access/vortex-tokio-local-disk 2254363 2.82066e+06 0.799231 ns
random-access/parquet-tokio-local-disk 233198812 2.43644e+08 0.957129 ns

@github-actions GitHub Actions
Copy link
Contributor

github-actions bot commented Feb 26, 2025 •
edited
Loading

Benchmarks: TPC-H on S3

Table of Results
name PR bfd7ebc base b219016 ratio (PR/base) unit
tpch_q01/parquet 272495586 251795772 1.08221 ns
tpch_q02/parquet 673640913 665037561 1.01294 ns
tpch_q03/parquet 439900464 413809822 1.06305 ns
tpch_q04/parquet 228764737 228971274 0.999098 ns
tpch_q05/parquet 587232875 576962593 1.0178 ns
tpch_q06/parquet 181682130 183908066 0.987896 ns
tpch_q07/parquet 639136770 618863454 1.03276 ns
tpch_q08/parquet 807393566 771955250 1.04591 ns
tpch_q09/parquet 699416349 670133506 1.0437 ns
tpch_q10/parquet 551074152 511994449 1.07633 ns
tpch_q11/parquet 312303426 262665731 1.18898 ns
tpch_q12/parquet 280729393 271553345 1.03379 ns
tpch_q13/parquet 404933153 388898302 1.04123 ns
tpch_q14/parquet 250933143 251891397 0.996196 ns
tpch_q15/parquet 472018712 474607290 0.994546 ns
tpch_q16/parquet 283457412 259052906 1.09421 ns
tpch_q17/parquet 395709968 393557295 1.00547 ns
tpch_q18/parquet 572017809 532286841 1.07464 ns
tpch_q19/parquet 303663532 274725653 1.10533 ns
tpch_q20/parquet 522712285 491389649 1.06374 ns
tpch_q21/parquet 634719147 617048250 1.02864 ns
tpch_q22/parquet 269424674 269338439 1.00032 ns
tpch_q01/vortex-file-compressed 148724032 146931023 1.0122 ns
tpch_q02/vortex-file-compressed 393747267 410850458 0.958371 ns
tpch_q03/vortex-file-compressed 299819895 295770574 1.01369 ns
tpch_q04/vortex-file-compressed 206011130 202789533 1.01589 ns
tpch_q05/vortex-file-compressed 326968651 326737852 1.00071 ns
tpch_q06/vortex-file-compressed 117060001 114910974 1.0187 ns
tpch_q07/vortex-file-compressed 390520071 375021593 1.04133 ns
tpch_q08/vortex-file-compressed 472730023 452631757 1.0444 ns
tpch_q09/vortex-file-compressed 401483065 385339055 1.0419 ns
tpch_q10/vortex-file-compressed 404405772 383550542 1.05437 ns
tpch_q11/vortex-file-compressed 167382127 155928761 1.07345 ns
tpch_q12/vortex-file-compressed 221176591 215370645 1.02696 ns
tpch_q13/vortex-file-compressed 181937634 180040657 1.01054 ns
tpch_q14/vortex-file-compressed 145561001 136321572 1.06778 ns
tpch_q15/vortex-file-compressed 313609523 300765561 1.0427 ns
tpch_q16/vortex-file-compressed 188262306 176935723 1.06402 ns
tpch_q17/vortex-file-compressed 215218971 209906636 1.02531 ns
tpch_q18/vortex-file-compressed 288329415 268366585 1.07439 ns
tpch_q19/vortex-file-compressed 207082590 193328990 1.07114 ns
tpch_q20/vortex-file-compressed 345435286 323877576 1.06656 ns
tpch_q21/vortex-file-compressed 489633761 469592629 1.04268 ns
tpch_q22/vortex-file-compressed 156341661 141245042 1.10688 ns

@github-actions GitHub Actions
Copy link
Contributor

github-actions bot commented Feb 26, 2025 •
edited
Loading

Benchmarks: TPC-H on NVME

Table of Results
name PR bfd7ebc base b219016 ratio (PR/base) unit
tpch_q01/arrow 43998318 41445469 1.0616 ns
tpch_q02/arrow 49835900 47749448 1.0437 ns
tpch_q03/arrow 31408098 28979094 1.08382 ns
tpch_q04/arrow 23901107 21051576 1.13536 ns
tpch_q05/arrow 50031159 46510524 1.0757 ns
tpch_q06/arrow 10205290 8354462 1.22154 ns
tpch_q07/arrow 78411279 72673027 1.07896 ns
tpch_q08/arrow 57963243 54915112 1.05551 ns
tpch_q09/arrow 72017515 66517283 1.08269 ns
tpch_q10/arrow 49147547 41755850 1.17702 ns
tpch_q11/arrow 25556917 23128159 1.10501 ns
tpch_q12/arrow 29983035 24619012 1.21788 ns
tpch_q13/arrow 17542449 14605459 1.20109 ns
tpch_q14/arrow 15795730 13420881 1.17695 ns
tpch_q15/arrow 30870879 24611050 1.25435 ns
tpch_q16/arrow 23216415 20169133 1.15109 ns
tpch_q17/arrow 65343870 58219368 1.12237 ns
tpch_q18/arrow 100238596 94017968 1.06616 ns
tpch_q19/arrow 29312270 24539984 1.19447 ns
tpch_q20/arrow 36220791 32559218 1.11246 ns
tpch_q21/arrow 117931172 108190737 1.09003 ns
tpch_q22/arrow 16149777 13463171 1.19955 ns
tpch_q01/parquet 113274262 111236442 1.01832 ns
tpch_q02/parquet 118372551 107163302 1.1046 ns
tpch_q03/parquet 105442328 101481361 1.03903 ns
tpch_q04/parquet 58992908 56292531 1.04797 ns
tpch_q05/parquet 118662122 114925283 1.03252 ns
tpch_q06/parquet 26698538 24712052 1.08039 ns
tpch_q07/parquet 138874046 129245017 1.0745 ns
tpch_q08/parquet 161805639 156743967 1.03229 ns
tpch_q09/parquet 213716805 202616504 1.05478 ns
tpch_q10/parquet 135349209 124801455 1.08452 ns
tpch_q11/parquet 56090799 50037172 1.12098 ns
tpch_q12/parquet 94630186 89472473 1.05765 ns
tpch_q13/parquet 151623856 145033936 1.04544 ns
tpch_q14/parquet 45228659 43500773 1.03972 ns
tpch_q15/parquet 74591417 62233733 1.19857 ns
tpch_q16/parquet 52039600 47877018 1.08694 ns
tpch_q17/parquet 139159507 125865059 1.10562 ns
tpch_q18/parquet 200965460 181216284 1.10898 ns
tpch_q19/parquet 74841741 72509083 1.03217 ns
tpch_q20/parquet 99804839 96697644 1.03213 ns
tpch_q21/parquet 189198791 173122983 1.09286 ns
tpch_q22/parquet 50816884 47164365 1.07744 ns
tpch_q01/vortex-file-compressed 36467117 34374898 1.06086 ns
tpch_q02/vortex-file-compressed 61426686 55267941 1.11143 ns
tpch_q03/vortex-file-compressed 34010416 29441298 1.15519 ns
tpch_q04/vortex-file-compressed 20937882 18505629 1.13143 ns
tpch_q05/vortex-file-compressed 49959730 44454222 1.12385 ns
tpch_q06/vortex-file-compressed 10217916 9356980 1.09201 ns
tpch_q07/vortex-file-compressed 74065604 66531394 1.11324 ns
tpch_q08/vortex-file-compressed 60506136 51724332 1.16978 ns
tpch_q09/vortex-file-compressed 71585000 64091059 1.11693 ns
tpch_q10/vortex-file-compressed 56439635 52369767 1.07771 ns
tpch_q11/vortex-file-compressed 25084898 23331648 1.07514 ns
tpch_q12/vortex-file-compressed 27155804 24499472 1.10842 ns
tpch_q13/vortex-file-compressed 24708265 24117325 1.0245 ns
tpch_q14/vortex-file-compressed 17229523 15151764 1.13713 ns
tpch_q15/vortex-file-compressed 31694796 27334273 1.15953 ns
tpch_q16/vortex-file-compressed 31140002 27117911 1.14832 ns
tpch_q17/vortex-file-compressed 54318520 49345209 1.10079 ns
tpch_q18/vortex-file-compressed 94245731 81621384 1.15467 ns
tpch_q19/vortex-file-compressed 37302764 30286117 1.23168 ns
tpch_q20/vortex-file-compressed 40340189 35217884 1.14545 ns
tpch_q21/vortex-file-compressed 95032648 85821445 1.10733 ns
tpch_q22/vortex-file-compressed 30573640 27261661 1.12149 ns

@github-actions GitHub Actions
Copy link
Contributor

github-actions bot commented Feb 26, 2025 •
edited
Loading

Benchmarks: compress

Table of Results
name PR 8b420cf base dffeeb8 ratio (PR/base) unit
compress time/taxi throughput 0.223259 0.213525 1.04559 bytes/ns
parquet_rs-zstd compress time/taxi throughput 0.281972 0.275991 1.02167 bytes/ns
decompress time/taxi throughput 1.90431 1.92364 0.989951 bytes/ns
parquet_rs-zstd decompress time/taxi throughput 1.66478 1.64399 1.01265 bytes/ns
compress time/AirlineSentiment throughput 0.00277229 0.0027646 1.00278 bytes/ns
parquet_rs-zstd compress time/AirlineSentiment throughput 0.0543871 0.0513611 1.05892 bytes/ns
decompress time/AirlineSentiment throughput 0.0274608 0.025891 1.06063 bytes/ns
parquet_rs-zstd decompress time/AirlineSentiment throughput 0.0906794 0.102908 0.881167 bytes/ns
compress time/Arade throughput 0.15937 0.157153 1.01411 bytes/ns
parquet_rs-zstd compress time/Arade throughput 0.398245 0.393812 1.01126 bytes/ns
decompress time/Arade throughput 1.87581 1.85093 1.01344 bytes/ns
parquet_rs-zstd decompress time/Arade throughput 1.92723 1.88557 1.0221 bytes/ns
compress time/Bimbo throughput 0.383936 0.365445 1.0506 bytes/ns
parquet_rs-zstd compress time/Bimbo throughput 0.335391 0.326937 1.02586 bytes/ns
decompress time/Bimbo throughput 2.20343 2.20185 1.00072 bytes/ns
parquet_rs-zstd decompress time/Bimbo throughput 2.80494 2.76901 1.01298 bytes/ns
compress time/CMSprovider throughput 0.0506614 0.0535887 0.945374 bytes/ns
parquet_rs-zstd compress time/CMSprovider throughput 0.351077 0.353393 0.993448 bytes/ns
decompress time/CMSprovider throughput 4.00317 4.07612 0.982103 bytes/ns
parquet_rs-zstd decompress time/CMSprovider throughput 1.79826 1.78508 1.00738 bytes/ns
compress time/Euro2016 throughput 0.155374 0.154625 1.00484 bytes/ns
parquet_rs-zstd compress time/Euro2016 throughput 0.302725 0.299284 1.0115 bytes/ns
decompress time/Euro2016 throughput 2.43934 2.54213 0.959567 bytes/ns
parquet_rs-zstd decompress time/Euro2016 throughput 0.999457 0.995602 1.00387 bytes/ns
compress time/Food throughput 0.190328 0.184741 1.03024 bytes/ns
parquet_rs-zstd compress time/Food throughput 0.316722 0.313067 1.01167 bytes/ns
decompress time/Food throughput 5.2472 5.18618 1.01176 bytes/ns
parquet_rs-zstd decompress time/Food throughput 1.6208 1.60187 1.01182 bytes/ns
compress time/HashTags throughput 0.19306 0.191881 1.00614 bytes/ns
parquet_rs-zstd compress time/HashTags throughput 0.809605 0.788074 1.02732 bytes/ns
decompress time/HashTags throughput 5.86061 5.79372 1.01155 bytes/ns
parquet_rs-zstd decompress time/HashTags throughput 2.79485 2.59373 1.07754 bytes/ns
compress time/TPC-H l_comment chunked throughput 0.217801 0.208791 1.04315 bytes/ns
parquet_rs-zstd compress time/TPC-H l_comment chunked throughput 0.285559 0.277746 1.02813 bytes/ns
decompress time/TPC-H l_comment chunked throughput 3.09212 3.03214 1.01978 bytes/ns
parquet_rs-zstd decompress time/TPC-H l_comment chunked throughput 1.05759 1.0636 0.994351 bytes/ns
compress time/TPC-H l_comment canonical throughput 0.0271484 0.028362 0.95721 bytes/ns
parquet_rs-zstd compress time/TPC-H l_comment canonical throughput 0.286067 0.281179 1.01738 bytes/ns
decompress time/TPC-H l_comment canonical throughput 3.10137 3.15698 0.982387 bytes/ns
parquet_rs-zstd decompress time/TPC-H l_comment canonical throughput 1.07627 1.05025 1.02477 bytes/ns
compress time/wide table cols=10 chunks=1 rows=1000 throughput 0.120606 0.128001 0.94223 bytes/ns
parquet_rs-zstd compress time/wide table cols=10 chunks=1 rows=1000 throughput 0.181194 0.181452 0.998582 bytes/ns
decompress time/wide table cols=10 chunks=1 rows=1000 throughput 0.732692 0.622132 1.17771 bytes/ns
parquet_rs-zstd decompress time/wide table cols=10 chunks=1 rows=1000 throughput 0.475465 0.515869 0.921676 bytes/ns
compress time/wide table cols=100 chunks=1 rows=1000 throughput 0.120016 0.127274 0.942977 bytes/ns
parquet_rs-zstd compress time/wide table cols=100 chunks=1 rows=1000 throughput 0.195428 0.181696 1.07558 bytes/ns
decompress time/wide table cols=100 chunks=1 rows=1000 throughput 1.06175 1.10568 0.96027 bytes/ns
parquet_rs-zstd decompress time/wide table cols=100 chunks=1 rows=1000 throughput 0.498996 0.504599 0.988896 bytes/ns
compress time/wide table cols=1000 chunks=1 rows=1000 throughput 0.109319 0.116863 0.935448 bytes/ns
parquet_rs-zstd compress time/wide table cols=1000 chunks=1 rows=1000 throughput 0.16931 0.168818 1.00292 bytes/ns
decompress time/wide table cols=1000 chunks=1 rows=1000 throughput 0.848971 0.825598 1.02831 bytes/ns
parquet_rs-zstd decompress time/wide table cols=1000 chunks=1 rows=1000 throughput 0.455769 0.448721 1.01571 bytes/ns
compress time/wide table cols=10 chunks=50 rows=1000 throughput 0.0503496 0.0735901 0.684189 bytes/ns
parquet_rs-zstd compress time/wide table cols=10 chunks=50 rows=1000 throughput 0.131767 0.125877 1.04679 bytes/ns
decompress time/wide table cols=10 chunks=50 rows=1000 throughput 0.832106 0.877309 0.948476 bytes/ns
parquet_rs-zstd decompress time/wide table cols=10 chunks=50 rows=1000 throughput 0.529507 0.541026 0.97871 bytes/ns
compress time/wide table cols=100 chunks=50 rows=1000 throughput 0.0495447 0.0677259 0.731548 bytes/ns
parquet_rs-zstd compress time/wide table cols=100 chunks=50 rows=1000 throughput 0.133699 0.115987 1.15271 bytes/ns
decompress time/wide table cols=100 chunks=50 rows=1000 throughput 1.07806 1.1413 0.944587 bytes/ns
parquet_rs-zstd decompress time/wide table cols=100 chunks=50 rows=1000 throughput 0.512486 0.526451 0.973473 bytes/ns
compress time/wide table cols=1000 chunks=50 rows=1000 throughput 0.0415182 0.0605364 0.685839 bytes/ns
parquet_rs-zstd compress time/wide table cols=1000 chunks=50 rows=1000 throughput 0.101328 0.0981726 1.03215 bytes/ns
decompress time/wide table cols=1000 chunks=50 rows=1000 throughput 0.898782 0.926083 0.970519 bytes/ns
parquet_rs-zstd decompress time/wide table cols=1000 chunks=50 rows=1000 throughput 0.45219 0.478746 0.94453 bytes/ns
vortex:raw size/taxi 0.117729 0.117729 1
vortex size/taxi 5.82471e+07 5.82471e+07 1
vortex:parquet-zstd size/taxi 1.04085 1.04085 1
vortex:raw size/AirlineSentiment 1.25903 1.25903 1
vortex size/AirlineSentiment 4112 4112 1
vortex:parquet-zstd size/AirlineSentiment 4.25233 4.25233 1
vortex:raw size/Arade 0.255852 0.255852 1
vortex size/Arade 3.03615e+08 3.03615e+08 1
vortex:parquet-zstd size/Arade 0.994178 0.994178 1
vortex:raw size/Bimbo 0.115538 0.115538 1
vortex size/Bimbo 8.25997e+08 8.25997e+08 1
vortex:parquet-zstd size/Bimbo 2.12803 2.12803 1
vortex:raw size/CMSprovider 0.188928 0.188928 1
vortex size/CMSprovider 1.18655e+09 1.18655e+09 1
vortex:parquet-zstd size/CMSprovider 1.54193 1.54193 1
vortex:raw size/Euro2016 0.471339 0.471345 0.999986
vortex size/Euro2016 2.1447e+08 2.14473e+08 0.999986
vortex:parquet-zstd size/Euro2016 1.80394 1.80396 0.999986
vortex:raw size/Food 0.177321 0.177321 1
vortex size/Food 5.97293e+07 5.97293e+07 1
vortex:parquet-zstd size/Food 1.64861 1.64861 1
vortex:raw size/HashTags 0.142738 0.142738 1
vortex size/HashTags 2.73374e+08 2.73374e+08 1
vortex:parquet-zstd size/HashTags 2.03913 2.03913 1
vortex:raw size/TPC-H l_comment chunked 0.418263 0.417913 1.00084
vortex size/TPC-H l_comment chunked 1.0423e+08 1.04143e+08 1.00084
vortex:parquet-zstd size/TPC-H l_comment chunked 1.83066 1.82926 1.00077
vortex:raw size/TPC-H l_comment canonical 0.425506 0.425404 1.00024
vortex size/TPC-H l_comment canonical 1.06032e+08 1.06007e+08 1.00024
vortex:parquet-zstd size/TPC-H l_comment canonical 1.86245 1.86183 1.00033
vortex:raw size/wide table cols=10 chunks=1 rows=1000 0.622482 0.622482 1
vortex size/wide table cols=10 chunks=1 rows=1000 99688 99688 1
vortex:parquet-zstd size/wide table cols=10 chunks=1 rows=1000 1.06636 1.06636 1
vortex:raw size/wide table cols=100 chunks=1 rows=1000 0.620919 0.620919 1
vortex size/wide table cols=100 chunks=1 rows=1000 994288 994288 1
vortex:parquet-zstd size/wide table cols=100 chunks=1 rows=1000 1.06363 1.06363 1
vortex:raw size/wide table cols=1000 chunks=1 rows=1000 0.620763 0.620763 1
vortex size/wide table cols=1000 chunks=1 rows=1000 9.94029e+06 9.94029e+06 1
vortex:parquet-zstd size/wide table cols=1000 chunks=1 rows=1000 1.06336 1.06336 1
vortex:raw size/wide table cols=10 chunks=50 rows=1000 0.597263 0.597263 1
vortex size/wide table cols=10 chunks=50 rows=1000 99688 99688 1
vortex:parquet-zstd size/wide table cols=10 chunks=50 rows=1000 1.06636 1.06636 1
vortex:raw size/wide table cols=100 chunks=50 rows=1000 0.597024 0.597024 1
vortex size/wide table cols=100 chunks=50 rows=1000 994288 994288 1
vortex:parquet-zstd size/wide table cols=100 chunks=50 rows=1000 1.06363 1.06363 1
vortex:raw size/wide table cols=1000 chunks=50 rows=1000 0.597 0.597 1
vortex size/wide table cols=1000 chunks=50 rows=1000 9.94029e+06 9.94029e+06 1
vortex:parquet-zstd size/wide table cols=1000 chunks=50 rows=1000 1.06336 1.06336 1

@robert3005 robert3005 force-pushed the dk/chunked-array-into-canonical-canonical-into-2 branch from 4035433 to 2d57fca Compare March 6, 2025 18:23
@robert3005
Copy link
Member

@danking @gatesn I have rebased this pr and made list and struct dtype not be recursive

@danking danking added the benchmark Run benchmarks on this branch label Mar 6, 2025
@danking
Copy link
Member Author

danking commented Mar 6, 2025

🧟

@github-actions github-actions bot removed the benchmark Run benchmarks on this branch label Mar 6, 2025
@github-actions GitHub Actions
Copy link
Contributor

github-actions bot commented Mar 6, 2025 •
edited
Loading

Benchmarks: Clickbench on NVME

Table of Results
name PR 4e147b2 base 590c413 ratio (PR/base) unit
clickbench_q00/parquet 2201274 2.46163e+06 0.894235 ns
clickbench_q01/parquet 33908458 3.11474e+07 1.08865 ns
clickbench_q02/parquet 66921331 6.1963e+07 1.08002 ns
clickbench_q03/parquet 54290967 5.08026e+07 1.06867 ns
clickbench_q04/parquet 331501412 3.03316e+08 1.09292 ns
clickbench_q05/parquet 313314632 2.94382e+08 1.06431 ns
clickbench_q06/parquet 1887919 2.30354e+06 0.819571 ns
clickbench_q07/parquet 32842041 3.13745e+07 1.04677 ns
clickbench_q08/parquet 386289755 3.64969e+08 1.05842 ns
clickbench_q09/parquet 574763190 5.36035e+08 1.07225 ns
clickbench_q10/parquet 119441969 1.11639e+08 1.0699 ns
clickbench_q11/parquet 140601339 1.3517e+08 1.04018 ns
clickbench_q12/parquet 315466317 3.00894e+08 1.04843 ns
clickbench_q13/parquet 486150909 4.58776e+08 1.05967 ns
clickbench_q14/parquet 323846475 3.03767e+08 1.0661 ns
clickbench_q15/parquet 351110224 3.4184e+08 1.02712 ns
clickbench_q16/parquet 745337931 7.39723e+08 1.00759 ns
clickbench_q17/parquet 691183583 6.38369e+08 1.08273 ns
clickbench_q18/parquet 1584497347 1.49463e+09 1.06012 ns
clickbench_q19/parquet 43416247 3.99775e+07 1.08602 ns
clickbench_q20/parquet 567920588 5.2527e+08 1.0812 ns
clickbench_q21/parquet 617688969 6.09546e+08 1.01336 ns
clickbench_q22/parquet 977115832 9.22323e+08 1.05941 ns
clickbench_q23/parquet 3915226010 3.69577e+09 1.05938 ns
clickbench_q24/parquet 196975832 1.8275e+08 1.07784 ns
clickbench_q25/parquet 171461905 1.61239e+08 1.0634 ns
clickbench_q26/parquet 218620723 2.0607e+08 1.0609 ns
clickbench_q27/parquet 755316714 7.11813e+08 1.06112 ns
clickbench_q28/parquet 4451513650 4.14999e+09 1.07266 ns
clickbench_q29/parquet 254876238 2.33725e+08 1.0905 ns
clickbench_q30/parquet 327758475 3.10847e+08 1.0544 ns
clickbench_q31/parquet 381224308 3.55963e+08 1.07097 ns
clickbench_q32/parquet 1872624986 1.7588e+09 1.06472 ns
clickbench_q33/parquet 1499204345 1.46961e+09 1.02013 ns
clickbench_q34/parquet 1484310010 1.43205e+09 1.0365 ns
clickbench_q35/parquet 510484128 4.86946e+08 1.04834 ns
clickbench_q36/parquet 146897131 1.42492e+08 1.03091 ns
clickbench_q37/parquet 68620007 6.39827e+07 1.07248 ns
clickbench_q38/parquet 93853569 8.97205e+07 1.04607 ns
clickbench_q39/parquet 282249989 2.70175e+08 1.04469 ns
clickbench_q40/parquet 45496531 4.29045e+07 1.06041 ns
clickbench_q41/parquet 42531240 4.12096e+07 1.03207 ns
clickbench_q42/parquet 56754558 5.09657e+07 1.11358 ns
clickbench_q00/vortex-file-compressed 4069814 3.96859e+06 1.02551 ns
clickbench_q01/vortex-file-compressed 14881589 1.45143e+07 1.02531 ns
clickbench_q02/vortex-file-compressed 27911228 2.66694e+07 1.04656 ns
clickbench_q03/vortex-file-compressed 39186402 3.59457e+07 1.09016 ns
clickbench_q04/vortex-file-compressed 339967497 3.29912e+08 1.03048 ns
clickbench_q05/vortex-file-compressed 322798344 3.22507e+08 1.0009 ns
clickbench_q06/vortex-file-compressed 4559938 4.02405e+06 1.13317 ns
clickbench_q07/vortex-file-compressed 17411821 1.72854e+07 1.00731 ns
clickbench_q08/vortex-file-compressed 403244028 3.71656e+08 1.08499 ns
clickbench_q09/vortex-file-compressed 482041025 4.56185e+08 1.05668 ns
clickbench_q10/vortex-file-compressed 63908467 6.10863e+07 1.0462 ns
clickbench_q11/vortex-file-compressed 70964724 6.84318e+07 1.03701 ns
clickbench_q12/vortex-file-compressed 254936992 2.45874e+08 1.03686 ns
clickbench_q13/vortex-file-compressed 361451038 3.37806e+08 1.07 ns
clickbench_q14/vortex-file-compressed 252009717 2.40862e+08 1.04628 ns
clickbench_q15/vortex-file-compressed 414336090 3.80475e+08 1.089 ns
clickbench_q16/vortex-file-compressed 790523249 7.46148e+08 1.05947 ns
clickbench_q17/vortex-file-compressed 777178470 7.38743e+08 1.05203 ns
clickbench_q18/vortex-file-compressed 1282206947 1.24054e+09 1.03358 ns
clickbench_q19/vortex-file-compressed 26336693 2.48946e+07 1.05793 ns
clickbench_q20/vortex-file-compressed 448499121 4.24246e+08 1.05717 ns
clickbench_q21/vortex-file-compressed 466356086 4.39311e+08 1.06156 ns
clickbench_q22/vortex-file-compressed 615462681 6.01514e+08 1.02319 ns
clickbench_q23/vortex-file-compressed 1234784394 1.19407e+09 1.0341 ns
clickbench_q24/vortex-file-compressed 87274871 7.9318e+07 1.10032 ns
clickbench_q25/vortex-file-compressed 98263513 9.20108e+07 1.06796 ns
clickbench_q26/vortex-file-compressed 115366777 1.08503e+08 1.06326 ns
clickbench_q27/vortex-file-compressed 700695208 6.61765e+08 1.05883 ns
clickbench_q28/vortex-file-compressed 4857061682 4.88988e+09 0.993289 ns
clickbench_q29/vortex-file-compressed 266409360 2.40578e+08 1.10737 ns
clickbench_q30/vortex-file-compressed 222537362 2.20949e+08 1.00719 ns
clickbench_q31/vortex-file-compressed 234471104 2.22688e+08 1.05291 ns
clickbench_q32/vortex-file-compressed 1341180150 1.24597e+09 1.07642 ns
clickbench_q33/vortex-file-compressed 1340678622 1.28734e+09 1.04143 ns
clickbench_q34/vortex-file-compressed 1342920116 1.28347e+09 1.04632 ns
clickbench_q35/vortex-file-compressed 606887934 5.7486e+08 1.05571 ns
clickbench_q36/vortex-file-compressed 60921294 5.36623e+07 1.13527 ns
clickbench_q37/vortex-file-compressed 37854695 3.30074e+07 1.14685 ns
clickbench_q38/vortex-file-compressed 26181144 2.3982e+07 1.0917 ns
clickbench_q39/vortex-file-compressed 109165798 1.0528e+08 1.03691 ns
clickbench_q40/vortex-file-compressed 24674606 1.98978e+07 1.24007 ns
clickbench_q41/vortex-file-compressed 21375346 1.87571e+07 1.13959 ns
clickbench_q42/vortex-file-compressed 30615420 2.7645e+07 1.10745 ns

@robert3005 robert3005 force-pushed the dk/chunked-array-into-canonical-canonical-into-2 branch from 2d57fca to 5be2f29 Compare March 7, 2025 10:49
@danking danking force-pushed the dk/chunked-array-into-canonical-canonical-into-2 branch from 5be2f29 to 91cf1f9 Compare March 7, 2025 14:22
@robert3005
Copy link
Member

eh @danking you wiped the changes I have made to preserve the old behaviour

@robert3005 robert3005 force-pushed the dk/chunked-array-into-canonical-canonical-into-2 branch from 91cf1f9 to 5be2f29 Compare March 7, 2025 15:14
@robert3005 robert3005 force-pushed the dk/chunked-array-into-canonical-canonical-into-2 branch from 5be2f29 to 21db3bf Compare March 7, 2025 15:14
@danking
Copy link
Member Author

danking commented Mar 7, 2025

whoops truly a sin that -f is --force and not --force-with-lease.

@danking
Copy link
Member Author

danking commented Mar 7, 2025

I don't understand why the SQL benchmarks aren't running

@danking
Copy link
Member Author

danking commented Mar 7, 2025 •
edited
Loading

maybe now working?

@robert3005
Copy link
Member

they're running

@danking
Copy link
Member Author

danking commented Mar 7, 2025

Hmm. It seems to have no macro effect though most microbenchmarks improve.

@robert3005
Copy link
Member

that's expected imho at this point. We are mostly bottlenecked on io and datafusion logic

@robert3005
Copy link
Member

but this is the last piece of builder canonical migration so would be nice to have it done

@danking danking merged commit 250c31b into develop Mar 7, 2025
32 of 33 checks passed
@danking danking deleted the dk/chunked-array-into-canonical-canonical-into-2 branch March 7, 2025 16:13
@robert3005 robert3005 mentioned this pull request Mar 12, 2025
Closed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@robert3005 robert3005 robert3005 approved these changes

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@danking @gatesn @robert3005