Open
Conversation
Signed-off-by: Nicholas Gates <nick@nickgates.com>
Signed-off-by: Nicholas Gates <nick@nickgates.com>
Signed-off-by: Nicholas Gates <nick@nickgates.com>
Contributor
|
I think you want this idea: vortex/vortex-array/src/arrays/dict/array.rs Line 144 in 74d803b |
Signed-off-by: Nicholas Gates <nick@nickgates.com>
Signed-off-by: Nicholas Gates <nick@nickgates.com> # Conflicts: # vortex-array/src/arrays/dict/vtable/mod.rs # vortex-duckdb/src/exporter/dict.rs
Merging this PR will improve performance by 24.89%
|
| Mode | Benchmark | BASE |
HEAD |
Efficiency | |
|---|---|---|---|---|---|
| ⚡ | Simulation | bench_many_codes_few_values[1024] |
397.5 µs | 325.1 µs | +22.27% |
| ⚡ | Simulation | bench_many_codes_few_values[2048] |
368.5 µs | 295.9 µs | +24.55% |
| ⚡ | Simulation | bench_many_codes_few_values[4096] |
374.3 µs | 301.7 µs | +24.08% |
| ⚡ | Simulation | bench_many_nulls[0.5] |
361.8 µs | 317.5 µs | +13.96% |
| ⚡ | Simulation | bench_many_nulls[0.9] |
537.1 µs | 456.3 µs | +17.72% |
| ⚡ | Simulation | bench_sparse_coverage[0.01] |
366.4 µs | 293.7 µs | +24.75% |
| ⚡ | Simulation | bench_sparse_coverage[0.1] |
364.8 µs | 292.1 µs | +24.89% |
| ⚡ | Simulation | bench_sparse_coverage[0.5] |
364.8 µs | 292.1 µs | +24.89% |
Comparing ngates/sparse-dict (22be09b) with develop (f3d5f09)
Signed-off-by: Nicholas Gates <nick@nickgates.com>
Signed-off-by: Nicholas Gates <nick@nickgates.com>
Signed-off-by: Nicholas Gates <nick@nickgates.com>
Signed-off-by: Nicholas Gates <nick@nickgates.com>
Signed-off-by: Nicholas Gates <nick@nickgates.com>
Signed-off-by: Nicholas Gates <nick@nickgates.com>
Signed-off-by: Nicholas Gates <nick@nickgates.com>
Signed-off-by: Nicholas Gates <nick@nickgates.com>
github-actions
Bot
removed
the
action/benchmark-sql
Contributor
Polar Signals Profiling ResultsLatest Run
Powered by Polar Signals Cloud |
Contributor
Benchmarks: PolarSignals ProfilingVortex (geomean): 1.005x ➖ datafusion / vortex-file-compressed (1.005x ➖, 0↑ 0↓)
|
Contributor
File Sizes: PolarSignals ProfilingNo file size changes detected. |
Contributor
Benchmarks: FineWeb NVMeVerdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (0.990x ➖, 0↑ 0↓)
datafusion / vortex-compact (1.010x ➖, 0↑ 0↓)
datafusion / parquet (0.982x ➖, 1↑ 0↓)
duckdb / vortex-file-compressed (1.040x ➖, 0↑ 1↓)
duckdb / vortex-compact (1.008x ➖, 0↑ 0↓)
duckdb / parquet (0.987x ➖, 1↑ 1↓)
Full attributed analysis
|
Contributor
File Sizes: FineWeb NVMeNo file size changes detected. |
Signed-off-by: Nicholas Gates <nick@nickgates.com>
Contributor
Benchmarks: TPC-H SF=1 on NVMEVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (1.009x ➖, 0↑ 0↓)
datafusion / vortex-compact (1.010x ➖, 0↑ 1↓)
datafusion / parquet (0.998x ➖, 2↑ 1↓)
datafusion / arrow (1.018x ➖, 0↑ 2↓)
duckdb / vortex-file-compressed (1.141x ❌, 0↑ 16↓)
duckdb / vortex-compact (1.067x ➖, 0↑ 1↓)
duckdb / parquet (1.098x ➖, 0↑ 12↓)
duckdb / duckdb (1.035x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: TPC-H SF=1 on NVMENo file size changes detected. |
Contributor
Benchmarks: TPC-DS SF=1 on NVMEVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (1.030x ➖, 0↑ 3↓)
datafusion / vortex-compact (1.030x ➖, 2↑ 4↓)
datafusion / parquet (1.020x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (1.067x ➖, 0↑ 26↓)
duckdb / vortex-compact (1.093x ➖, 0↑ 40↓)
duckdb / parquet (1.019x ➖, 1↑ 5↓)
duckdb / duckdb (1.059x ➖, 0↑ 18↓)
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 (1.021x ➖, 0↑ 1↓)
datafusion / vortex-compact (1.045x ➖, 0↑ 0↓)
datafusion / parquet (0.984x ➖, 0↑ 1↓)
duckdb / vortex-file-compressed (1.001x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.037x ➖, 0↑ 0↓)
duckdb / parquet (1.017x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
Benchmarks: Statistical and Population GeneticsVerdict: No clear signal (low confidence) duckdb / vortex-file-compressed (0.979x ➖, 1↑ 0↓)
duckdb / vortex-compact (0.979x ➖, 1↑ 0↓)
duckdb / parquet (1.007x ➖, 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 (0.996x ➖, 0↑ 0↓)
datafusion / vortex-compact (1.011x ➖, 0↑ 0↓)
datafusion / parquet (0.965x ➖, 0↑ 0↓)
datafusion / arrow (0.971x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (0.986x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.994x ➖, 0↑ 0↓)
duckdb / parquet (1.008x ➖, 0↑ 0↓)
duckdb / duckdb (0.992x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: TPC-H SF=10 on NVMENo file size changes detected. |
Contributor
Benchmarks: TPC-H SF=1 on S3Verdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (1.019x ➖, 1↑ 0↓)
datafusion / vortex-compact (0.989x ➖, 1↑ 1↓)
datafusion / parquet (1.115x ➖, 0↑ 4↓)
duckdb / vortex-file-compressed (1.056x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.090x ➖, 0↑ 0↓)
duckdb / parquet (1.025x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
Benchmarks: Clickbench on NVMEVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (1.075x ➖, 0↑ 13↓)
datafusion / parquet (1.069x ➖, 0↑ 10↓)
duckdb / vortex-file-compressed (1.121x ❌, 0↑ 26↓)
duckdb / parquet (1.038x ➖, 0↑ 3↓)
duckdb / duckdb (1.045x ➖, 0↑ 3↓)
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=10 on S3Verdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (1.100x ➖, 0↑ 4↓)
datafusion / vortex-compact (1.011x ➖, 1↑ 0↓)
datafusion / parquet (1.020x ➖, 0↑ 1↓)
duckdb / vortex-file-compressed (1.000x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.016x ➖, 0↑ 0↓)
duckdb / parquet (1.099x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
Author
|
The problem is, we now run a rather expensive check on every canonicalize... Need to think about how we can constrain this some more. I think it may involve storing the approximate unique count of codes as an array statistic |
robert3005
requested changes
May 8, 2026
Contributor
robert3005
left a comment
robert3005
left a comment
There was a problem hiding this comment.
I think beyond the exporter changes that are unclear to me why they're in this pr the rest of the code is fine, I mostly left some questions since it looks like too much llm code
Comment on lines
24
to
+44
| // TODO(joe): if the range is size 1 replace with a constant array | ||
| if let Some(code) = sliced_code.as_opt::<Constant>() { | ||
| let code = code.scalar().as_primitive().as_::<usize>(); | ||
| return if let Some(code) = code { | ||
| let values = array.values().slice(code..code + 1)?; | ||
| Ok(Some( | ||
| DictArray::new( | ||
| // SAFETY: the only dictionary value is referenced by every non-null code when the | ||
| // slice is non-empty. An empty code stream cannot reference a non-empty values | ||
| // array. | ||
| let sliced = unsafe { | ||
| DictArray::new_unchecked( | ||
| ConstantArray::new(0u8, sliced_code.len()).into_array(), | ||
| values, | ||
| ) | ||
| .into_array(), | ||
| )) | ||
| }; | ||
| if sliced_code.is_empty() { | ||
| Ok(Some(sliced.into_array())) | ||
| } else { | ||
| Ok(Some(unsafe { | ||
| sliced.set_all_values_referenced(true).into_array() | ||
| })) | ||
| } |
Contributor
There was a problem hiding this comment.
can't you return values in this code path? Am I missing something very obvious?
| } | ||
|
|
||
| if !array.has_all_values_referenced() | ||
| // `take(FilterArray(...))` is also represented as a dictionary. Compacting that shape |
| pub struct ConversionCache { | ||
| pub dict_cache: DashMap<usize, (ArrayRef, ReusableDict)>, | ||
| pub values_cache: DashMap<usize, (ArrayRef, Arc<Mutex<Vector>>)>, | ||
| pub canonical_cache: DashMap<usize, (ArrayRef, Canonical)>, |
Contributor
There was a problem hiding this comment.
little bit confused why this would be removed here
| use crate::executor::ExecutionCtx; | ||
| use crate::validity::Validity; | ||
|
|
||
| // TODO: Replace this fixed sparse-dictionary threshold with a cost model that accounts for values |
Contributor
There was a problem hiding this comment.
this should be an issue number
| array: &DictArray, | ||
| ctx: &mut ExecutionCtx, | ||
| ) -> VortexResult<Option<Canonical>> { | ||
| let codes = array.codes().as_::<Primitive>().into_owned(); |
Contributor
There was a problem hiding this comment.
this looks wrong, and will panic in practice
Contributor
There was a problem hiding this comment.
This will be fine but it's unclear from here that we have previously canonicalzed codes
| ctx: &mut ExecutionCtx, | ||
| ) -> VortexResult<Option<Canonical>> { | ||
| let codes = array.codes().as_::<Primitive>().into_owned(); | ||
| let Some(sparse_codes) = collect_sparse_codes(&codes, array.values().len(), ctx)? else { |
Contributor
There was a problem hiding this comment.
It's weird that you're passing things by reference here if you made the value owned line before
| // Sampling is only a preflight for cases that are not sparse by row count alone. Keep it away | ||
| // from tiny dictionary domains and near-dense slices where the estimator overhead dominates. | ||
| values_len >= SPARSE_CANONICALIZE_MIN_SAMPLED_VALUES_LEN | ||
| && codes_len.saturating_mul(SPARSE_CANONICALIZE_SAMPLED_CODES_PER_VALUE_THRESHOLD) |
Contributor
There was a problem hiding this comment.
part of me would want to remove SPARSE_CANONICALIZE prefix so they're different
Comment on lines
+217
to
+218
| unique_codes: PrimitiveArray::new(Buffer::from_iter(unique_codes), Validity::NonNullable), | ||
| remapped_codes: PrimitiveArray::new(Buffer::from_iter(remapped_codes), validity), |
Contributor
There was a problem hiding this comment.
why are you copying here
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.
When the number of unique codes is much much smaller than the number of values, we should collect unique codes, filter the values, and remap the codes into the smaller dictionary's value space.
I don't expect this PR to immediately have any impact. But I have some other changes that propagate filter masks deeper into the tree so they actually hit the DictLayout