gpu compatible write strategy, move compact strategy to use btrblocks with zstd and pco #6322
Conversation
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
CodSpeed Performance ReportMerging this PR will not alter performanceComparing Summary
Footnotes
|
Comment on lines
222
to
207
| /// Configure a write strategy that emits only CUDA-compatible encodings. | ||
| /// | ||
| /// This keeps the default write layout pipeline, but: | ||
| /// - Restricts flat-layout normalization to [`GPU_ALLOWED_ENCODINGS`] | ||
| /// - Configures BtrBlocks to exclude schemes without CUDA kernel support | ||
| #[cfg(feature = "zstd")] | ||
| pub fn with_cuda_compatible_encodings(mut self) -> Self { | ||
| let btrblocks = BtrBlocksCompressorBuilder::default() | ||
| .exclude_int([IntCode::Sparse, IntCode::Rle]) | ||
| .exclude_float([FloatCode::AlpRd, FloatCode::Rle, FloatCode::Sparse]) | ||
| // Keep string schemes disabled in btrblocks; when `zstd` feature is enabled, we | ||
| // separately encode string/binary leaves as Zstd (without dictionaries). | ||
| .exclude_string([ | ||
| StringCode::Dict, | ||
| StringCode::Fsst, | ||
| StringCode::Constant, | ||
| StringCode::Sparse, | ||
| ]) | ||
| .build(); | ||
|
|
||
| self.compressor = Some(Arc::new(GpuCompatibleCompressor::new(btrblocks))); | ||
| self.allow_encodings = Some((*GPU_ALLOWED_ENCODINGS).clone()); | ||
| self | ||
| } |
Contributor
There was a problem hiding this comment.
this should be done by the caller?
Contributor
Author
There was a problem hiding this comment.
it can, but now compact also doesn't use a custom compressor so it is useful to keep these as convenience imo
Contributor
There was a problem hiding this comment.
its the cfg zstd new (is that the cuda zstd) or regular
onursatici
changed the title
onursatici
added
the
action/benchmark-sql
github-actions
bot
removed
the
action/benchmark-sql
Contributor
Benchmarks: PolarSignals ProfilingSummary
Detailed Results Table
|
Contributor
Benchmarks: TPC-H SF=1 on NVMESummary
Detailed Results Table
|
Contributor
Benchmarks: FineWeb NVMeSummary
Detailed Results Table
|
Contributor
Benchmarks: TPC-DS SF=1 on NVMESummary
Detailed Results Table
|
Contributor
Benchmarks: TPC-H SF=1 on S3Summary
Detailed Results Table
|
Contributor
Benchmarks: TPC-H SF=10 on NVMESummary
Detailed Results Table
|
Contributor
Benchmarks: FineWeb S3Summary
Detailed Results Table
|
Contributor
Benchmarks: Statistical and Population GeneticsSummary
Detailed Results Table
|
Contributor
Benchmarks: TPC-H SF=10 on S3Summary
Detailed Results Table
|
Contributor
Benchmarks: Clickbench on NVMESummary
Detailed Results Table
|
| &SequenceScheme, | ||
| &RLE_INTEGER_SCHEME, | ||
| #[cfg(feature = "pco")] | ||
| &PcoScheme, |
Contributor
There was a problem hiding this comment.
not must not be in the default
joseph-isaacs
requested changes
Feb 6, 2026
Contributor
joseph-isaacs
left a comment
joseph-isaacs
left a comment
There was a problem hiding this comment.
make sure teh default strategy doesn't have PCO or zstd
Contributor
Author
|
default is filtering out pco and zstd |
onursatici
commented
Feb 6, 2026
| int_schemes: ALL_INT_SCHEMES | ||
| .iter() | ||
| .copied() | ||
| .filter(|s| s.code() != IntCode::Pco) |
Contributor
Author
There was a problem hiding this comment.
here is currently where we exclude pco and zstd from the default
onursatici
added
the
action/benchmark-sql
github-actions
bot
removed
the
action/benchmark-sql
Comment on lines
+54
to
+57
| int_schemes: ALL_INT_SCHEMES | ||
| .iter() | ||
| .copied() | ||
| .filter(|s| s.code() != IntCode::Pco) |
Contributor
There was a problem hiding this comment.
shall we have ALL_DEFAULT_SCEHEMS? next to all schemes?
Contributor
There was a problem hiding this comment.
its just people will forget to omit other schemes
joseph-isaacs
approved these changes
Feb 6, 2026
Contributor
joseph-isaacs
left a comment
joseph-isaacs
left a comment
There was a problem hiding this comment.
Add the default schemes and we can merge
joseph-isaacs
requested changes
Feb 6, 2026
Contributor
joseph-isaacs
left a comment
joseph-isaacs
left a comment
There was a problem hiding this comment.
need to fix the python test, this shouldn't change
File "../vortex-python/python/vortex/io.py", line ?, in default
Failed example:
os.path.getsize('tiny.vortex')
Expected:
55116
Got:
55316
joseph-isaacs
approved these changes
Feb 6, 2026
Signed-off-by: Onur Satici <onur@spiraldb.com>
Signed-off-by: Onur Satici <onur@spiraldb.com>
onursatici
force-pushed
the
os/gpu-strategy
branch
from
f4f3811 to
f475bdb
Compare
onursatici
added
the
action/benchmark-sql
github-actions
bot
removed
the
action/benchmark-sql
Contributor
Author
|
the test failure is because the tiny vortex file in the test doesn't accommodate for the fastlanes 1024 byte padding that would be included. We should be accommodating that on larger samples because we sample with windows that are multiples of 1024, but for small arrays FoR compression estimation would overshoot, causing the compact compressor to pick FoR instead of Pco. That is why the test file is 200 byte larger, it did pick FoR instead of Pco for the zones child, and with the padding FoR ended up being larger that the primitive array so we fell back to primitive |
Contributor
|
Just noticed this change because of a conflict - I think this design is a bit suboptimal since Pco+Zstd will ~always be the strongest encodings for their respective data types, so here we're 1. adding a small risk that sampling will randomly choose something weaker based a small portion of the data (e.g. I think @onursatici's last comment is exactly this problem) and 2. increasing compression time somewhat. I think instead of simply adding encodings, we should have different sets depending on the compression strength the user wants. |
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.
No description provided.