mbutrovich opened a new pull request, #18875:
URL: https://github.com/apache/datafusion/pull/18875
## 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 #18487.
## 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.
-->
DataFusion Comet often uses Sort Merge Joins because DataFusion does have a
larger-than-memory Hash Join operator. Performance on TPC-H Q21 is quite bad
when run through native, and instead Comet falls back to Spark by default. If
you force Comet to use DataFusion's SMJ operator, performance is:
<img width="969" height="618" alt="Screenshot 2025-11-21 at 11 31 18 AM"
src="https://github.com/user-attachments/assets/bb04d9c4-4f05-4ab8-8960-47a4705d2c83";
/>
Profiling showed most of the time spent in `concat_batches` of single-digit
rows:
<img width="1132" height="348" alt="Screenshot 2025-11-20 at 6 49 20 PM"
src="https://github.com/user-attachments/assets/89ef70d2-41e5-46cd-82b2-d97b2af1de56";
/>
## 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.
-->
Use a `BatchCoalescer` both internally and to buffer final output. One made
the biggest difference, but I switched to two to be consistent. Here are Comet
results with the changes based on 50.3 (which is where Comet is):
<img width="952" height="599" alt="Screenshot 2025-11-21 at 11 43 57 AM"
src="https://github.com/user-attachments/assets/0660addd-acd8-41ed-8b1f-d9efd31e2a17";
/>
I ported this PR to DF 51 afterwards, and found
https://github.com/apache/datafusion/pull/17193 changed the API around batch
coalescing. I might need to update this to support `LimitedBatchCoalescer`like
https://github.com/apache/datafusion/pull/18783
Benchmark results:
```
./bench.sh compare_detail main smj
Comparing main and smj
--------------------
Benchmark smj.json
--------------------
┏━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query ┃ main ┃
smj ┃ Change ┃
┡━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 1 │ 2.04 / 3.34 ±2.56 / 8.45 ms │ 2.00 / 3.29
±2.45 / 8.19 ms │ no change │
│ QQuery 2 │ 12.91 / 13.47 ±0.79 / 15.03 ms │ 12.42 / 14.09
±1.79 / 17.46 ms │ no change │
│ QQuery 3 │ 68.83 / 73.67 ±3.59 / 78.93 ms │ 74.27 / 76.41
±1.69 / 78.58 ms │ no change │
│ QQuery 4 │ 2.24 / 2.78 ±0.71 / 4.18 ms │ 2.17 / 2.87
±0.75 / 4.33 ms │ no change │
│ QQuery 5 │ 20.46 / 22.31 ±1.18 / 23.99 ms │ 18.85 / 19.68
±0.59 / 20.47 ms │ +1.13x faster │
│ QQuery 6 │ 12.25 / 12.97 ±0.57 / 13.60 ms │ 12.55 / 12.89
±0.31 / 13.38 ms │ no change │
│ QQuery 7 │ 15.21 / 15.70 ±0.35 / 16.18 ms │ 16.77 / 17.24
±0.30 / 17.62 ms │ 1.10x slower │
│ QQuery 8 │ 2.47 / 2.59 ±0.11 / 2.78 ms │ 2.46 / 2.62
±0.09 / 2.75 ms │ no change │
│ QQuery 9 │ 15.91 / 16.21 ±0.37 / 16.91 ms │ 16.23 / 16.90
±0.68 / 18.20 ms │ no change │
│ QQuery 10 │ 5.69 / 5.77 ±0.12 / 6.00 ms │ 5.64 / 5.98
±0.23 / 6.31 ms │ no change │
│ QQuery 11 │ 3.36 / 3.49 ±0.11 / 3.69 ms │ 2.79 / 3.08
±0.19 / 3.26 ms │ +1.13x faster │
│ QQuery 12 │ 79.53 / 81.43 ±2.38 / 86.11 ms │ 30.25 / 30.66
±0.54 / 31.71 ms │ +2.66x faster │
│ QQuery 13 │ 5.60 / 5.78 ±0.13 / 5.99 ms │ 5.64 / 5.79
±0.15 / 6.01 ms │ no change │
│ QQuery 14 │ 5.66 / 5.95 ±0.17 / 6.16 ms │ 5.60 / 5.80
±0.25 / 6.28 ms │ no change │
│ QQuery 15 │ 5.48 / 5.68 ±0.15 / 5.91 ms │ 5.44 / 5.81
±0.29 / 6.17 ms │ no change │
│ QQuery 16 │ 1.73 / 1.84 ±0.07 / 1.92 ms │ 1.80 / 1.85
±0.04 / 1.90 ms │ no change │
│ QQuery 17 │ 10.75 / 11.08 ±0.23 / 11.46 ms │ 9.56 / 10.59
±0.56 / 11.26 ms │ no change │
│ QQuery 18 │ 9.09 / 9.67 ±0.35 / 10.02 ms │ 7.97 / 8.27
±0.26 / 8.61 ms │ +1.17x faster │
│ QQuery 19 │ 24.72 / 25.67 ±0.72 / 26.59 ms │ 24.84 / 25.74
±0.68 / 26.81 ms │ no change │
│ QQuery 20 │ 861.69 / 909.57 ±39.61 / 958.08 ms │ 915.42 / 971.44 ±37.08
/ 1030.81 ms │ 1.07x slower │
└──────────────┴────────────────────────────────────┴─────────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┓
┃ Benchmark Summary ┃ ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━┩
│ Total Time (main) │ 1228.97ms │
│ Total Time (smj) │ 1241.00ms │
│ Average Time (main) │ 61.45ms │
│ Average Time (smj) │ 62.05ms │
│ Queries Faster │ 4 │
│ Queries Slower │ 2 │
│ Queries with No Change │ 14 │
│ Queries with Failure │ 0 │
└────────────────────────┴───────────┘
--------------------
Benchmark tpch_sf1.json
--------------------
┏━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━┓
┃ Query ┃ main ┃
smj ┃ Change ┃
┡━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━┩
│ QQuery 1 │ 44.37 / 48.67 ±4.54 / 55.68 ms │ 49.10 /
51.78 ±2.55 / 56.50 ms │ 1.06x slower │
│ QQuery 2 │ 45.18 / 47.44 ±2.39 / 51.74 ms │ 45.98 /
47.07 ±1.10 / 49.15 ms │ no change │
│ QQuery 3 │ 52.59 / 56.15 ±2.65 / 59.79 ms │ 55.38 /
60.31 ±5.21 / 66.90 ms │ 1.07x slower │
│ QQuery 4 │ 33.06 / 34.46 ±0.97 / 35.88 ms │ 32.39 /
33.70 ±0.80 / 34.47 ms │ no change │
│ QQuery 5 │ 84.50 / 87.63 ±2.06 / 90.58 ms │ 84.21 /
87.19 ±2.37 / 90.40 ms │ no change │
│ QQuery 6 │ 17.87 / 18.64 ±0.48 / 19.22 ms │ 16.78 /
18.06 ±0.96 / 19.37 ms │ no change │
│ QQuery 7 │ 111.11 / 113.59 ±1.79 / 116.70 ms │ 108.28 /
116.71 ±7.15 / 127.92 ms │ no change │
│ QQuery 8 │ 89.84 / 94.59 ±3.34 / 100.15 ms │ 96.64 /
101.78 ±3.39 / 106.00 ms │ 1.08x slower │
│ QQuery 9 │ 128.36 / 133.12 ±3.46 / 138.00 ms │ 130.89 /
140.20 ±8.09 / 150.47 ms │ 1.05x slower │
│ QQuery 10 │ 49.89 / 51.91 ±1.41 / 54.19 ms │ 47.69 /
51.04 ±5.55 / 62.09 ms │ no change │
│ QQuery 11 │ 34.19 / 35.30 ±0.59 / 35.84 ms │ 32.05 /
33.28 ±0.86 / 34.42 ms │ +1.06x faster │
│ QQuery 12 │ 36.26 / 38.67 ±2.44 / 42.77 ms │ 32.70 /
33.54 ±0.55 / 34.16 ms │ +1.15x faster │
│ QQuery 13 │ 31.32 / 34.13 ±2.29 / 38.22 ms │ 28.54 /
29.94 ±1.64 / 33.14 ms │ +1.14x faster │
│ QQuery 14 │ 23.54 / 24.79 ±0.92 / 26.00 ms │ 22.30 /
23.21 ±0.84 / 24.71 ms │ +1.07x faster │
│ QQuery 15 │ 26.66 / 27.47 ±0.86 / 29.05 ms │ 25.81 /
27.60 ±1.14 / 29.37 ms │ no change │
│ QQuery 16 │ 17.63 / 18.94 ±0.97 / 20.20 ms │ 16.97 /
17.95 ±0.57 / 18.57 ms │ +1.06x faster │
│ QQuery 17 │ 94.36 / 96.41 ±1.62 / 98.44 ms │ 92.20 /
94.14 ±2.07 / 98.03 ms │ no change │
│ QQuery 18 │ 99.91 / 108.58 ±5.85 / 117.27 ms │ 103.51 /
107.20 ±2.69 / 111.14 ms │ no change │
│ QQuery 19 │ 35.23 / 36.68 ±1.46 / 39.23 ms │ 32.54 /
34.29 ±1.04 / 35.46 ms │ +1.07x faster │
│ QQuery 20 │ 40.66 / 41.84 ±1.20 / 44.05 ms │ 37.83 /
38.51 ±0.63 / 39.46 ms │ +1.09x faster │
│ QQuery 21 │ 151142.04 / 246274.24 ±89682.07 / 358766.84 ms │ 209.69 /
210.83 ±1.09 / 212.70 ms │ +1168.11x faster │
│ QQuery 22 │ 16.69 / 28.53 ±22.72 / 73.97 ms │ 16.37 /
16.89 ±0.31 / 17.35 ms │ +1.69x faster │
└──────────────┴────────────────────────────────────────────────┴───────────────────────────────────┴──────────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━┓
┃ Benchmark Summary ┃ ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━┩
│ Total Time (main) │ 247451.79ms │
│ Total Time (smj) │ 1375.22ms │
│ Average Time (main) │ 11247.81ms │
│ Average Time (smj) │ 62.51ms │
│ Queries Faster │ 9 │
│ Queries Slower │ 4 │
│ Queries with No Change │ 9 │
│ Queries with Failure │ 0 │
└────────────────────────┴─────────────┘
```
## 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)?
-->
Existing Sort Merge Join unit tests, added a new benchmark.
## Are there any user-facing changes?
<!--
If there are user-facing changes then we may require documentation to be
updated before approving the PR.
-->
There should not be.
<!--
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: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
