perf: remove implicit ListViewArray rebuild during take and filter operations#8048
Open
mhk197 wants to merge 13 commits into
Open
perf: remove implicit ListViewArray rebuild during take and filter operations#8048mhk197 wants to merge 13 commits into
ListViewArray rebuild during take and filter operations#8048mhk197 wants to merge 13 commits into
Conversation
github-actions
Bot
removed
the
action/benchmark
mhk197
changed the title
ListViewArray rebuild during take and filter operationsListViewArray rebuild during take and filter operations
mhk197
force-pushed
the
mk/explicit-listview-compaction
branch
from
c6a1940 to
95ae305
Compare
Contributor
Polar Signals Profiling ResultsLatest Run
Powered by Polar Signals Cloud |
Merging this PR will improve performance by 19.67%
|
| Mode | Benchmark | BASE |
HEAD |
Efficiency | |
|---|---|---|---|---|---|
| ⚡ | Simulation | chunked_varbinview_opt_canonical_into[(1000, 10)] |
225.1 µs | 188.1 µs | +19.67% |
Tip
Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.
Comparing mk/explicit-listview-compaction (cd7c97e) with develop (f852d72)
Contributor
Benchmarks: PolarSignals ProfilingVortex (geomean): 1.052x ➖ datafusion / vortex-file-compressed (1.052x ➖, 0↑ 2↓)
|
Contributor
File Sizes: PolarSignals ProfilingFile Size Changes (1 files changed, -0.0% overall, 0↑ 1↓)
Totals:
|
Contributor
Benchmarks: FineWeb NVMeVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (0.927x ➖, 2↑ 0↓)
datafusion / vortex-compact (0.928x ➖, 3↑ 0↓)
datafusion / parquet (0.926x ➖, 2↑ 0↓)
duckdb / vortex-file-compressed (0.925x ➖, 3↑ 0↓)
duckdb / vortex-compact (0.972x ➖, 1↑ 0↓)
duckdb / parquet (0.936x ➖, 1↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: FineWeb NVMeFile Size Changes (2 files changed, -0.0% overall, 0↑ 2↓)
Totals:
|
Contributor
Benchmarks: TPC-H SF=1 on NVMEVerdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (1.004x ➖, 0↑ 0↓)
datafusion / vortex-compact (0.999x ➖, 0↑ 0↓)
datafusion / parquet (1.002x ➖, 1↑ 1↓)
datafusion / arrow (1.003x ➖, 0↑ 1↓)
duckdb / vortex-file-compressed (1.005x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.003x ➖, 0↑ 0↓)
duckdb / parquet (1.013x ➖, 1↑ 4↓)
duckdb / duckdb (1.001x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: TPC-H SF=1 on NVMEFile Size Changes (18 files changed, -0.0% overall, 0↑ 18↓)
Totals:
|
Contributor
Benchmarks: TPC-DS SF=1 on NVMEVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (0.989x ➖, 2↑ 0↓)
datafusion / vortex-compact (0.992x ➖, 2↑ 1↓)
datafusion / parquet (1.008x ➖, 0↑ 3↓)
duckdb / vortex-file-compressed (0.998x ➖, 1↑ 1↓)
duckdb / vortex-compact (0.995x ➖, 1↑ 0↓)
duckdb / parquet (0.996x ➖, 0↑ 1↓)
duckdb / duckdb (1.000x ➖, 2↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: TPC-DS SF=1 on NVMEFile Size Changes (48 files changed, -0.0% overall, 0↑ 48↓)
Totals:
|
Contributor
Benchmarks: FineWeb S3Verdict: No clear signal (low confidence) datafusion / vortex-file-compressed (1.084x ➖, 0↑ 1↓)
datafusion / vortex-compact (0.985x ➖, 0↑ 0↓)
datafusion / parquet (1.009x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (1.007x ➖, 0↑ 1↓)
duckdb / vortex-compact (0.990x ➖, 0↑ 0↓)
duckdb / parquet (0.988x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
Benchmarks: Statistical and Population GeneticsVerdict: No clear signal (low confidence) duckdb / vortex-file-compressed (0.995x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.010x ➖, 0↑ 1↓)
duckdb / parquet (0.990x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: Statistical and Population GeneticsFile Size Changes (2 files changed, -0.0% overall, 0↑ 2↓)
Totals:
|
Contributor
Benchmarks: Random AccessVortex (geomean): 0.847x ✅ unknown / unknown (0.882x ✅, 19↑ 0↓)
|
Contributor
Benchmarks: TPC-H SF=10 on NVMEVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (1.047x ➖, 0↑ 7↓)
datafusion / vortex-compact (1.056x ➖, 0↑ 8↓)
datafusion / parquet (0.994x ➖, 0↑ 3↓)
datafusion / arrow (0.979x ➖, 1↑ 3↓)
duckdb / vortex-file-compressed (1.001x ➖, 0↑ 3↓)
duckdb / vortex-compact (0.975x ➖, 0↑ 0↓)
duckdb / parquet (0.994x ➖, 0↑ 0↓)
duckdb / duckdb (0.975x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: TPC-H SF=10 on NVMEFile Size Changes (48 files changed, -0.0% overall, 0↑ 48↓)
Totals:
|
Contributor
Benchmarks: Clickbench on NVMEVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (1.003x ➖, 0↑ 0↓)
datafusion / parquet (0.990x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (1.000x ➖, 1↑ 2↓)
duckdb / parquet (0.998x ➖, 0↑ 0↓)
duckdb / duckdb (0.984x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: Clickbench on NVMEFile Size Changes (201 files changed, -0.0% overall, 0↑ 201↓)
Totals:
|
Contributor
Benchmarks: TPC-H SF=1 on S3Verdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (0.995x ➖, 2↑ 4↓)
datafusion / vortex-compact (0.878x ➖, 3↑ 0↓)
datafusion / parquet (0.995x ➖, 0↑ 2↓)
duckdb / vortex-file-compressed (1.031x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.075x ➖, 0↑ 0↓)
duckdb / parquet (1.106x ➖, 0↑ 1↓)
Full attributed analysis
|
Contributor
Benchmarks: CompressionVortex (geomean): 1.000x ➖ unknown / unknown (0.999x ➖, 7↑ 3↓)
|
Contributor
Benchmarks: TPC-H SF=10 on S3Verdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (1.012x ➖, 0↑ 2↓)
datafusion / vortex-compact (0.935x ➖, 1↑ 0↓)
datafusion / parquet (1.004x ➖, 1↑ 1↓)
duckdb / vortex-file-compressed (0.975x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.976x ➖, 0↑ 0↓)
duckdb / parquet (1.025x ➖, 0↑ 0↓)
Full attributed analysis
|
mhk197
force-pushed
the
mk/explicit-listview-compaction
branch
from
f8b4f2e to
933bdc1
Compare
connortsui20
requested changes
May 21, 2026
mhk197
force-pushed
the
mk/explicit-listview-compaction
branch
from
ad614a7 to
7490746
Compare
Signed-off-by: Matthew Katz <katz@spiraldb.com>
Signed-off-by: Matthew Katz <katz@spiraldb.com>
Signed-off-by: Matthew Katz <katz@spiraldb.com>
Signed-off-by: Matthew Katz <katz@spiraldb.com>
Signed-off-by: Matthew Katz <katz@spiraldb.com>
Signed-off-by: Matthew Katz <katz@spiraldb.com>
Signed-off-by: Matthew Katz <katz@spiraldb.com>
Signed-off-by: Matthew Katz <katz@spiraldb.com>
Signed-off-by: Matthew Katz <katz@spiraldb.com>
Signed-off-by: Matthew Katz <katz@spiraldb.com>
mhk197
force-pushed
the
mk/explicit-listview-compaction
branch
from
57908ab to
cd7c97e
Compare
Contributor
joseph-isaacs
left a comment
joseph-isaacs
left a comment
There was a problem hiding this comment.
looks. good overall
did you do this for duckdb too? that is important. since that might be passed sparse element lists.
also a single definitive comment explaining what you did.
| .as_::<u64>() | ||
| .ok_or_else(|| vortex_err!("could not cast sum of sizes to u64"))?; | ||
|
|
||
| let estimate = (sizes_sum as f32 / n_elts as f32).clamp(0.0, 1.0); |
Contributor
There was a problem hiding this comment.
how can this go above or below 0,1 maybe a panic here.
or debug assert
Contributor
There was a problem hiding this comment.
I would prefer a debug assert here
Author
There was a problem hiding this comment.
Elements: [0,1]
ListView: [[0,1], [1]]
----
estimate: 3/2
Can't go below 0 though ofc, so I can use explicit > 1 check and change instead of clamp
| } | ||
| } | ||
|
|
||
| pub fn should_rebuild(&self, exact: bool, ctx: &mut ExecutionCtx) -> VortexResult<bool> { |
Contributor
There was a problem hiding this comment.
Like Joe said, this seems like a footgun. I think that the caller should just decide, and the REBUILD_DENSITY_THRESHOLD can just be DEFAULT_REBUILD_DENSITY_THRESHOLD or similar
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.
Summary
Removes implicit rebuilds from
ListViewArraytakeandfiltercompute kernels, adds density-introspection methods to support deciding when to rebuild explicitly at materialization boundaries, and defers rebuilding until array export.Motivation. The previous code rebuilt the elements buffer eagerly on every
takeorfilterwhose row-fraction dropped belowREBUILD_DENSITY_THRESHOLD. In an execution tree liketake → take → ..., an eager mid-pipeline rebuild costs an allocation and a full copy of referenced ranges that the next operator may immediately sparsify away.The row-fraction heuristic was also inaccurate. It doesn't account for per-row size variance, unreferenced elements, and duplicate references.
Changes:
TakeReduce::takeandTakeExecute::takefilter_listviewcompute_referenced_elements_mask,compute_density, andestimate_densityListViewArray::should_rebuild(exact: bool): Currently unused but intended to gate an explicit compaction boundary in follow-up.API Changes
take/filterover a sparse or low-densityListViewArrayno longer allocates a compacted elements buffer. Downstream consumers must callrebuildandshould_rebuildexplicitly themselves.Testing
vortex-array/src/arrays/listview/tests/density.rswith multiple tests: