You are viewing a plain text version of this content. The canonical link for it is here.
Posted to github@datafusion.apache.org by "Dandandan (via GitHub)" <gi...@apache.org> on 2026/03/08 12:54:46 UTC
[PR] Use u32 group indices for aggregations to reduce memory usage [datafusion]
Dandandan opened a new pull request, #20801:
URL: https://github.com/apache/datafusion/pull/20801
Introduce GroupIndex type alias (u32) for group indices in aggregations, replacing the previous usize. This halves the memory per group index on 64-bit platforms (4 bytes vs 8 bytes), improving cache utilization during hash aggregation. A u32 supports up to ~4 billion groups, which is more than sufficient in practice since accumulator state memory would be exhausted long before reaching that limit.
Changes:
- Add `GroupIndex = u32` type alias in `datafusion-expr-common`
- Update `GroupsAccumulator` trait methods to use `&[GroupIndex]`
- Update `GroupValues::intern()` to produce `Vec<GroupIndex>`
- Update all accumulate helpers (NullState, accumulate_indices, etc.)
- Update all GroupsAccumulator and GroupValues implementations
- Update FFI layer to use `RVec<u32>` for group indices
- Update call sites in row_hash, recursive_query, and order tracking
https://claude.ai/code/session_011CtJW17mfcKqhc5SYCs7JZ
## Which issue does this PR close?
<!--
We generally require a GitHub issue to be filed for all bug fixes and enhancements and this helps us generate change logs for our releases. You can link an issue to this PR using the GitHub syntax. For example `Closes #123` indicates that this PR will close issue #123.
-->
- Closes #.
## Rationale for this change
<!--
Why are you proposing this change? If this is already explained clearly in the issue then this section is not needed.
Explaining clearly why changes are proposed helps reviewers understand your changes and offer better suggestions for fixes.
-->
## What changes are included in this PR?
<!--
There is no need to duplicate the description in the issue here but it is sometimes worth providing a summary of the individual changes in this PR.
-->
## Are these changes tested?
<!--
We typically require tests for all PRs in order to:
1. Prevent the code from being accidentally broken by subsequent changes
2. Serve as another way to document the expected behavior of the code
If tests are not included in your PR, please explain why (for example, are they covered by existing tests)?
-->
## Are there any user-facing changes?
<!--
If there are user-facing changes then we may require documentation to be updated before approving the PR.
-->
<!--
If there are any breaking changes to public APIs, please add the `api change` label.
-->
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For additional commands, e-mail: github-help@datafusion.apache.org
Re: [PR] Use u32 group indices for aggregations to reduce memory usage [datafusion]
Posted by "Dandandan (via GitHub)" <gi...@apache.org>.
Dandandan commented on PR #20801:
URL: https://github.com/apache/datafusion/pull/20801#issuecomment-4019002319
Run benchmarks
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For additional commands, e-mail: github-help@datafusion.apache.org
Re: [PR] [Bench] u32 group indices [datafusion]
Posted by "alamb-ghbot (via GitHub)" <gi...@apache.org>.
alamb-ghbot commented on PR #20801:
URL: https://github.com/apache/datafusion/pull/20801#issuecomment-4023686497
๐ค `./gh_compare_branch.sh` [gh_compare_branch.sh](https://github.com/alamb/datafusion-benchmarking/blob/main/scripts/gh_compare_branch.sh) Running
Linux aal-dev 6.14.0-1018-gcp #19~24.04.1-Ubuntu SMP Wed Sep 24 23:23:09 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux
Comparing claude/optimize-aggregations-u32-5Vv7z (accb52f3e774b7a1631d4bf1de6f81b507b67fe3) to 92078d932df77699496ab226802d3fe50b40fed2 [diff](https://github.com/apache/datafusion/compare/92078d932df77699496ab226802d3fe50b40fed2..accb52f3e774b7a1631d4bf1de6f81b507b67fe3) using: tpch_mem clickbench_partitioned clickbench_extended
Results will be posted here when complete
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For additional commands, e-mail: github-help@datafusion.apache.org
Re: [PR] [Bench] u32 group indices [datafusion]
Posted by "Dandandan (via GitHub)" <gi...@apache.org>.
Dandandan closed pull request #20801: [Bench] u32 group indices
URL: https://github.com/apache/datafusion/pull/20801
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For additional commands, e-mail: github-help@datafusion.apache.org
Re: [PR] [Bench] u32 group indices [datafusion]
Posted by "alamb-ghbot (via GitHub)" <gi...@apache.org>.
alamb-ghbot commented on PR #20801:
URL: https://github.com/apache/datafusion/pull/20801#issuecomment-4023932759
๐ค: Benchmark completed
<details><summary>Details</summary>
<p>
```
Comparing HEAD and claude_optimize-aggregations-u32-5Vv7z
--------------------
Benchmark clickbench_extended.json
--------------------
โโโโโโโโโโโโณโโโโโโโโโโโโโโณโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโณโโโโโโโโโโโโโโโ
โ Query โ HEAD โ claude_optimize-aggregations-u32-5Vv7z โ Change โ
โกโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโฉ
โ QQuery 0 โ 2177.31 ms โ 2228.66 ms โ no change โ
โ QQuery 1 โ 756.65 ms โ 860.13 ms โ 1.14x slower โ
โ QQuery 2 โ 1615.14 ms โ 1735.07 ms โ 1.07x slower โ
โ QQuery 3 โ 1004.98 ms โ 1043.56 ms โ no change โ
โ QQuery 4 โ 2106.99 ms โ 2113.28 ms โ no change โ
โ QQuery 5 โ 26650.05 ms โ 26157.24 ms โ no change โ
โ QQuery 6 โ 3634.04 ms โ 3817.63 ms โ 1.05x slower โ
โ QQuery 7 โ 2484.80 ms โ 2409.37 ms โ no change โ
โโโโโโโโโโโโดโโโโโโโโโโโโโโดโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโดโโโโโโโโโโโโโโโ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโณโโโโโโโโโโโโโ
โ Benchmark Summary โ โ
โกโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโฉ
โ Total Time (HEAD) โ 40429.97ms โ
โ Total Time (claude_optimize-aggregations-u32-5Vv7z) โ 40364.96ms โ
โ Average Time (HEAD) โ 5053.75ms โ
โ Average Time (claude_optimize-aggregations-u32-5Vv7z) โ 5045.62ms โ
โ Queries Faster โ 0 โ
โ Queries Slower โ 3 โ
โ Queries with No Change โ 5 โ
โ Queries with Failure โ 0 โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโดโโโโโโโโโโโโโ
--------------------
Benchmark clickbench_partitioned.json
--------------------
โโโโโโโโโโโโโณโโโโโโโโโโโโโโณโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโณโโโโโโโโโโโโโโโโ
โ Query โ HEAD โ claude_optimize-aggregations-u32-5Vv7z โ Change โ
โกโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโฉ
โ QQuery 0 โ 2.55 ms โ 2.59 ms โ no change โ
โ QQuery 1 โ 46.81 ms โ 46.28 ms โ no change โ
โ QQuery 2 โ 151.92 ms โ 148.73 ms โ no change โ
โ QQuery 3 โ 155.82 ms โ 157.02 ms โ no change โ
โ QQuery 4 โ 961.14 ms โ 954.40 ms โ no change โ
โ QQuery 5 โ 1178.05 ms โ 1223.42 ms โ no change โ
โ QQuery 6 โ 6.26 ms โ 6.82 ms โ 1.09x slower โ
โ QQuery 7 โ 50.70 ms โ 53.96 ms โ 1.06x slower โ
โ QQuery 8 โ 1332.28 ms โ 1347.87 ms โ no change โ
โ QQuery 9 โ 1703.69 ms โ 1765.90 ms โ no change โ
โ QQuery 10 โ 313.38 ms โ 333.34 ms โ 1.06x slower โ
โ QQuery 11 โ 351.89 ms โ 383.82 ms โ 1.09x slower โ
โ QQuery 12 โ 1117.82 ms โ 1148.01 ms โ no change โ
โ QQuery 13 โ 1739.09 ms โ 1826.77 ms โ 1.05x slower โ
โ QQuery 14 โ 1103.14 ms โ 1162.27 ms โ 1.05x slower โ
โ QQuery 15 โ 1100.72 ms โ 1114.98 ms โ no change โ
โ QQuery 16 โ 2284.11 ms โ 2345.55 ms โ no change โ
โ QQuery 17 โ 2260.32 ms โ 2324.02 ms โ no change โ
โ QQuery 18 โ 4371.24 ms โ 4485.12 ms โ no change โ
โ QQuery 19 โ 115.24 ms โ 120.97 ms โ no change โ
โ QQuery 20 โ 1688.17 ms โ 1792.74 ms โ 1.06x slower โ
โ QQuery 21 โ 1924.29 ms โ 2061.46 ms โ 1.07x slower โ
โ QQuery 22 โ 3357.49 ms โ 3550.88 ms โ 1.06x slower โ
โ QQuery 23 โ 10828.24 ms โ 11467.30 ms โ 1.06x slower โ
โ QQuery 24 โ 177.50 ms โ 185.68 ms โ no change โ
โ QQuery 25 โ 408.09 ms โ 431.90 ms โ 1.06x slower โ
โ QQuery 26 โ 184.70 ms โ 196.32 ms โ 1.06x slower โ
โ QQuery 27 โ 2451.25 ms โ 2603.71 ms โ 1.06x slower โ
โ QQuery 28 โ 21602.52 ms โ 23100.32 ms โ 1.07x slower โ
โ QQuery 29 โ 956.28 ms โ 959.86 ms โ no change โ
โ QQuery 30 โ 1172.33 ms โ 1193.06 ms โ no change โ
โ QQuery 31 โ 1242.88 ms โ 1260.20 ms โ no change โ
โ QQuery 32 โ 4119.72 ms โ 3790.19 ms โ +1.09x faster โ
โ QQuery 33 โ 4978.08 ms โ 5170.86 ms โ no change โ
โ QQuery 34 โ 5874.89 ms โ 5776.21 ms โ no change โ
โ QQuery 35 โ 1087.81 ms โ 1084.90 ms โ no change โ
โ QQuery 36 โ 181.34 ms โ 184.35 ms โ no change โ
โ QQuery 37 โ 69.35 ms โ 72.42 ms โ no change โ
โ QQuery 38 โ 108.99 ms โ 116.39 ms โ 1.07x slower โ
โ QQuery 39 โ 318.49 ms โ 342.67 ms โ 1.08x slower โ
โ QQuery 40 โ 40.22 ms โ 41.56 ms โ no change โ
โ QQuery 41 โ 33.92 ms โ 34.20 ms โ no change โ
โ QQuery 42 โ 30.88 ms โ 31.10 ms โ no change โ
โโโโโโโโโโโโโดโโโโโโโโโโโโโโดโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโดโโโโโโโโโโโโโโโโ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโณโโโโโโโโโโโโโ
โ Benchmark Summary โ โ
โกโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโฉ
โ Total Time (HEAD) โ 83183.62ms โ
โ Total Time (claude_optimize-aggregations-u32-5Vv7z) โ 86400.13ms โ
โ Average Time (HEAD) โ 1934.50ms โ
โ Average Time (claude_optimize-aggregations-u32-5Vv7z) โ 2009.31ms โ
โ Queries Faster โ 1 โ
โ Queries Slower โ 16 โ
โ Queries with No Change โ 26 โ
โ Queries with Failure โ 0 โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโดโโโโโโโโโโโโโ
--------------------
Benchmark tpch_mem_sf1.json
--------------------
โโโโโโโโโโโโโณโโโโโโโโโโโโณโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโณโโโโโโโโโโโโโโโโ
โ Query โ HEAD โ claude_optimize-aggregations-u32-5Vv7z โ Change โ
โกโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโฉ
โ QQuery 1 โ 102.40 ms โ 102.26 ms โ no change โ
โ QQuery 2 โ 30.54 ms โ 31.09 ms โ no change โ
โ QQuery 3 โ 34.49 ms โ 35.38 ms โ no change โ
โ QQuery 4 โ 31.28 ms โ 30.31 ms โ no change โ
โ QQuery 5 โ 82.05 ms โ 81.91 ms โ no change โ
โ QQuery 6 โ 20.04 ms โ 20.41 ms โ no change โ
โ QQuery 7 โ 147.24 ms โ 142.23 ms โ no change โ
โ QQuery 8 โ 39.30 ms โ 39.45 ms โ no change โ
โ QQuery 9 โ 101.12 ms โ 98.91 ms โ no change โ
โ QQuery 10 โ 62.82 ms โ 62.76 ms โ no change โ
โ QQuery 11 โ 18.73 ms โ 18.50 ms โ no change โ
โ QQuery 12 โ 54.00 ms โ 53.21 ms โ no change โ
โ QQuery 13 โ 48.14 ms โ 48.32 ms โ no change โ
โ QQuery 14 โ 14.60 ms โ 14.62 ms โ no change โ
โ QQuery 15 โ 29.32 ms โ 29.15 ms โ no change โ
โ QQuery 16 โ 26.87 ms โ 26.79 ms โ no change โ
โ QQuery 17 โ 142.73 ms โ 138.86 ms โ no change โ
โ QQuery 18 โ 262.86 ms โ 266.19 ms โ no change โ
โ QQuery 19 โ 41.14 ms โ 41.90 ms โ no change โ
โ QQuery 20 โ 57.76 ms โ 52.87 ms โ +1.09x faster โ
โ QQuery 21 โ 192.01 ms โ 190.21 ms โ no change โ
โ QQuery 22 โ 22.29 ms โ 22.28 ms โ no change โ
โโโโโโโโโโโโโดโโโโโโโโโโโโดโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโดโโโโโโโโโโโโโโโโ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโณโโโโโโโโโโโโ
โ Benchmark Summary โ โ
โกโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโฉ
โ Total Time (HEAD) โ 1561.74ms โ
โ Total Time (claude_optimize-aggregations-u32-5Vv7z) โ 1547.64ms โ
โ Average Time (HEAD) โ 70.99ms โ
โ Average Time (claude_optimize-aggregations-u32-5Vv7z) โ 70.35ms โ
โ Queries Faster โ 1 โ
โ Queries Slower โ 0 โ
โ Queries with No Change โ 21 โ
โ Queries with Failure โ 0 โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโดโโโโโโโโโโโโ
```
</p>
</details>
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For additional commands, e-mail: github-help@datafusion.apache.org