Ungrouped aggregate function pushdown for duckdb#8645
Merged
Conversation
myrrc
added
changelog/performance
12 tasks
12 tasks
Contributor
Author
|
As we don't have statistics attached to arrays, and we don't use precomputed statistics as Datafusion does, this PR would likely sit idle for some time until these are done, because currently there's no performance improvement compared to the default versions. There will be significant improvement once we add zone map statistics |
myrrc
changed the title
github-actions
Bot
removed
the
action/benchmark
Contributor
Polar Signals Profiling ResultsLatest Run
Previous Runs (6)
Powered by Polar Signals Cloud |
Contributor
Benchmarks: PolarSignals ProfilingVortex (geomean): 0.947x ➖ How to read Verdict and Engines
datafusion / vortex-file-compressed (0.947x ➖, 1↑ 0↓)
No file size changes detected. |
Contributor
Benchmarks: TPC-H SF=1 on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.797x ✅, 22↑ 0↓)
datafusion / vortex-compact (0.801x ✅, 22↑ 0↓)
datafusion / parquet (0.840x ✅, 20↑ 0↓)
datafusion / arrow (0.739x ✅, 22↑ 0↓)
duckdb / vortex-file-compressed (0.838x ✅, 21↑ 0↓)
duckdb / vortex-compact (0.837x ✅, 21↑ 0↓)
duckdb / parquet (0.875x ✅, 12↑ 0↓)
duckdb / duckdb (0.869x ✅, 17↑ 0↓)
File Size Changes (9 files changed, -0.0% overall, 5↑ 4↓)
Totals:
|
Contributor
Benchmarks: TPC-DS SF=1 on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.968x ➖, 3↑ 0↓)
datafusion / vortex-compact (0.971x ➖, 3↑ 0↓)
datafusion / parquet (0.970x ➖, 4↑ 1↓)
duckdb / vortex-file-compressed (0.982x ➖, 1↑ 0↓)
duckdb / vortex-compact (0.991x ➖, 0↑ 1↓)
duckdb / parquet (0.988x ➖, 0↑ 0↓)
duckdb / duckdb (0.990x ➖, 2↑ 0↓)
File Size Changes (6 files changed, +0.0% overall, 4↑ 2↓)
Totals:
|
Contributor
Benchmarks: FineWeb NVMeVerdict: No clear signal (environment too noisy confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.954x ➖, 0↑ 0↓)
datafusion / vortex-compact (0.968x ➖, 1↑ 0↓)
datafusion / parquet (1.046x ➖, 0↑ 5↓)
duckdb / vortex-file-compressed (0.942x ➖, 1↑ 0↓)
duckdb / vortex-compact (0.990x ➖, 0↑ 0↓)
duckdb / parquet (0.965x ➖, 1↑ 0↓)
File Size Changes (1 files changed, -0.0% overall, 0↑ 1↓)
Totals:
|
Contributor
Benchmarks: Statistical and Population GeneticsVerdict: No clear signal (low confidence) How to read Verdict and Engines
duckdb / vortex-file-compressed (1.037x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.042x ➖, 0↑ 0↓)
duckdb / parquet (1.020x ➖, 0↑ 0↓)
File Size Changes (1 files changed, +0.0% overall, 1↑ 0↓)
Totals:
|
Contributor
Benchmarks: FineWeb S3Verdict: No clear signal (environment too noisy confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.937x ➖, 1↑ 0↓)
datafusion / vortex-compact (1.050x ➖, 0↑ 1↓)
datafusion / parquet (0.880x ➖, 1↑ 0↓)
duckdb / vortex-file-compressed (0.908x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.912x ➖, 0↑ 0↓)
duckdb / parquet (0.914x ➖, 0↑ 0↓)
|
Contributor
Benchmarks: Clickbench Sorted on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.011x ➖, 1↑ 1↓)
datafusion / parquet (0.994x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (1.014x ➖, 0↑ 1↓)
duckdb / parquet (0.990x ➖, 0↑ 0↓)
duckdb / duckdb (1.008x ➖, 0↑ 1↓)
File Size Changes (201 files changed, +0.0% overall, 119↑ 82↓)
Totals:
|
Contributor
Benchmarks: TPC-H SF=10 on NVMEVerdict: No clear signal (medium confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.954x ➖, 0↑ 0↓)
datafusion / vortex-compact (0.999x ➖, 0↑ 0↓)
datafusion / parquet (0.948x ➖, 0↑ 0↓)
datafusion / arrow (0.890x ✅, 15↑ 0↓)
duckdb / vortex-file-compressed (0.756x ✅, 22↑ 0↓)
duckdb / vortex-compact (0.790x ✅, 22↑ 0↓)
duckdb / parquet (0.933x ➖, 7↑ 0↓)
duckdb / duckdb (0.889x ✅, 12↑ 0↓)
File Size Changes (27 files changed, +0.0% overall, 16↑ 11↓)
Totals:
|
Merging this PR will not alter performance
|
| Mode | Benchmark | BASE |
HEAD |
Efficiency | |
|---|---|---|---|---|---|
| ❌ | Simulation | chunked_varbinview_into_canonical[(1000, 10)] |
169.7 µs | 206 µs | -17.61% |
| ❌ | Simulation | encode_varbinview[(10000, 2)] |
994.8 µs | 1,133.4 µs | -12.23% |
| ⚡ | Simulation | bitwise_not_vortex_buffer_mut[128] |
273.6 ns | 215.3 ns | +27.1% |
| ⚡ | Simulation | bitwise_not_vortex_buffer_mut[1024] |
333.9 ns | 275.6 ns | +21.17% |
| ⚡ | Simulation | bitwise_not_vortex_buffer_mut[2048] |
427.8 ns | 369.4 ns | +15.79% |
| ⚡ | Simulation | chunked_varbinview_into_canonical[(100, 100)] |
307.1 µs | 271.9 µs | +12.94% |
| ⚡ | Simulation | chunked_varbinview_opt_canonical_into[(100, 100)] |
340.5 µs | 305.4 µs | +11.51% |
Tip
Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.
Comparing myrrc/temp-agg-fn-pushdown (c5e7653) with develop (11f15a9)
Footnotes
-
42 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩
Contributor
Benchmarks: Clickbench on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.931x ➖, 8↑ 0↓)
datafusion / parquet (0.941x ➖, 4↑ 0↓)
duckdb / vortex-file-compressed (0.982x ➖, 2↑ 0↓)
duckdb / parquet (0.999x ➖, 0↑ 0↓)
duckdb / duckdb (1.008x ➖, 0↑ 2↓)
File Size Changes (104 files changed, +0.0% overall, 53↑ 51↓)
Totals:
|
Contributor
Benchmarks: Random AccessVortex (geomean): 1.032x ➖ How to read Verdict and Engines
unknown / unknown (1.056x ➖, 0↑ 3↓)
|
Contributor
Benchmarks: TPC-H SF=10 on S3Verdict: No clear signal (environment too noisy confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.073x ➖, 0↑ 1↓)
datafusion / vortex-compact (1.024x ➖, 0↑ 0↓)
datafusion / parquet (0.953x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (0.993x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.998x ➖, 0↑ 0↓)
duckdb / parquet (1.036x ➖, 0↑ 0↓)
|
Contributor
Benchmarks: TPC-H SF=1 on S3Verdict: No clear signal (environment too noisy confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.952x ➖, 0↑ 0↓)
datafusion / vortex-compact (0.898x ➖, 2↑ 0↓)
datafusion / parquet (1.172x ➖, 0↑ 7↓)
duckdb / vortex-file-compressed (0.977x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.968x ➖, 0↑ 0↓)
duckdb / parquet (0.991x ➖, 0↑ 0↓)
|
Contributor
Benchmarks: Appian on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.015x ➖, 0↑ 0↓)
datafusion / parquet (0.930x ➖, 1↑ 0↓)
duckdb / vortex-file-compressed (1.082x ➖, 0↑ 1↓)
duckdb / parquet (1.063x ➖, 0↑ 0↓)
duckdb / duckdb (1.110x ❌, 0↑ 6↓)
File Size Changes (3 files changed, -0.0% overall, 1↑ 2↓)
Totals:
|
Contributor
Benchmarks: CompressionVortex (geomean): 0.993x ➖ How to read Verdict and Engines
unknown / unknown (0.956x ➖, 22↑ 2↓)
|
0ax1
reviewed
Jul 6, 2026
| let mut accumulators = global_state.partials.lock(); | ||
| let row_count = global_state.row_count.load(Ordering::Acquire) as i64; | ||
| let mut accum_iter = accumulators.iter_mut(); | ||
| for (i, aggregate) in global_state.aggregates.iter().enumerate() { |
Contributor
There was a problem hiding this comment.
This loop looks a bit manual? How large does accumulators get, should we do sth along the lines of zip / filter here and iter twice?
Contributor
Author
There was a problem hiding this comment.
We support 6 distinct aggregations but there may be many more columns for them.
Contributor
There was a problem hiding this comment.
Saying we could split it into 2 loops and make this simpler?
Merged
myrrc
force-pushed
the
myrrc/temp-agg-fn-pushdown
branch
from
afd1cd5 to
2b742eb
Compare
github-actions
Bot
removed
the
action/benchmark
myrrc
force-pushed
the
myrrc/temp-agg-fn-pushdown
branch
from
2b742eb to
10c6a50
Compare
myrrc
force-pushed
the
myrrc/temp-agg-fn-pushdown
branch
2 times, most recently
from
25b9ddb to
06cb841
Compare
myrrc
force-pushed
the
myrrc/temp-agg-fn-pushdown
branch
from
06cb841 to
f90f65d
Compare
github-actions
Bot
removed
the
action/benchmark
0ax1
reviewed
Jul 9, 2026
| let mut accumulators = global_state.partials.lock(); | ||
| let row_count = global_state.row_count.load(Ordering::Acquire) as i64; | ||
| let mut accum_iter = accumulators.iter_mut(); | ||
| for (i, aggregate) in global_state.aggregates.iter().enumerate() { |
Contributor
There was a problem hiding this comment.
Saying we could split it into 2 loops and make this simpler?
myrrc
force-pushed
the
myrrc/temp-agg-fn-pushdown
branch
from
f90f65d to
9d108d7
Compare
Signed-off-by: Mikhail Kot <mikhail@spiraldb.com>
myrrc
force-pushed
the
myrrc/temp-agg-fn-pushdown
branch
from
9d108d7 to
c5e7653
Compare
0ax1
approved these changes
Jul 9, 2026
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
2 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.
This PR adds support for pushing down ungropped aggregates min,max,sum,avg,mean,first,any_value,count(col), and count(*).
It consists of roughly 3 parts:
I've tried the implementation with CountStart being a separate accumulator, and it is much more complex than current one, because you can't provide a real column to accumulate on in select(), and you need to create a dummy one. So, we pay the cost of a separate atomic and a count_star positions vector, but our expression projection handling stays simple.
We also reject count(*) if it's the only aggregate because duckdb calculating it natively (just using chunk lengths) is faster than our accumulator pipeline
On the API input note, nearly all aggregate functions except for count_star() in duckdb have one non-const argument so input is a pair of a column and an expression.
Depends on #8649