Arrow Rust + Conbench Integration

[dianaclarke]

Here's a minimal Arrow Rust + Conbench[1] proof of concept.

[1] https://github.com/conbench/conbench

A few notes (areas for improvement, caveats, etc):

• Criterion results are in nanoseconds, but the smallest unit
  Conbench currently speaks is seconds (because Conbench was initially
  for macro not micro benchmarking). I suspect most places in Conbench
  would work just fine if nanoseconds were passed in, but I need to
  audit the code for any places that assume seconds if it isn't a
  throughput benchmark.

• If the Criterion benchmarks were named better, I could tag them
  better in Conbench. For example, I suspect sqrt_20_12, sqrt_20_9,
  sqrt_22_12, and sqrt_22_14 are parameterized variations of the same
  benchmark, and if they were named something like "sqrt, foo=20,
  bar=12", I could batch them together & tag their parameters so that
  Conbench would automatically graph them in relation to each other. I
  was sort of able to do this with the following benchmarks (because
  there was a machine readable pattern). Anyhoo, that's easy enough to
  do down the road as a last integration step, and it does appear from
  the Criterion docs that they have their own recommendations for how to
  do this.

  • window partition by, u64_narrow, aggregate functions
  • window partition by, u64_narrow, built-in functions
  • window partition by, u64_wide, aggregate functions
  • window partition by, u64_wide, built-in functions

• While Criterion benchmarks can also measure throughput in some
  cases, all the arrow-datafusion benchmarks were in elapsed time (not
  sure about the arrow-rs benchmarks), so I didn't bother writing code
  to support potential throughput results from
  arrow-datafusion/arrow-rs, but we may need to revisit that.

• We probably want to add some additional context, like the
  arrow-rs/arrow-datafusion version, rust version, any compiler flags,
  etc.

[codecov-commenter]

Codecov Report

Merging #1289 (6c3e7b8) into master (35e16be) will not change coverage.
The diff coverage is n/a.

@@ Coverage Diff @@## master #1289 +/- ##
=======================================
Coverage 83.04% 83.04% =======================================
Files 180 180 Lines 52424 52424 =======================================
Hits 43537 43537 Misses 8887 8887 

Impacted Files	Coverage Δ
arrow/src/datatypes/datatype.rs	66.40% <0.00%> (-0.40%)	⬇️
arrow/src/datatypes/field.rs	53.79% <0.00%> (-0.31%)	⬇️
parquet_derive/src/parquet_field.rs	66.21% <0.00%> (ø)
arrow/src/array/transform/mod.rs	84.77% <0.00%> (+0.25%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 35e16be...6c3e7b8. Read the comment docs.

[dianaclarke]

From a Python perspective, it's probably a mistake to name this directory conbench. Thoughts on what you would like arrow-rs/conbench/ to be named?

• arrow-rs/_conbench/?
• arrow-rs/conbench-benchmarks/?
• arrow-rs/conbench-integration/?

[andygrove]

I pulled the branch locally and followed the instructions in the README and benchmarks are currently running. It looks like it will take a while so I will check in on results in the morning and continue reviewing the PR.

[dianaclarke]

I pulled the branch locally and followed the instructions in the README and benchmarks are currently running. It looks like it will take a while so I will check in on results in the morning and continue reviewing the PR.

Nice!

99% of the time is spent executing cargo bench.

Note that this is only the first step – you should see json formatted benchmark results spew to your terminal, but they won't yet publish to the Arrow Conbench server[1].

You'll want to work with @ElenaHenderson on getting the results to publish to the Arrow Conbench server via buildkite etc.

[1] https://conbench.ursa.dev/

[dianaclarke]

This is just the GitHub generated .gitignore for Python packages. I can trim in down, but I usually don't bother.

[andygrove]

One of the tests failed because it makes an assumption that a target directory exists and it did not. My assumption is that somehow the current working directory was not correct. This is how I initiated the benchmark run.

~/git/arrow-rs/conbench$ conbench arrow-rs --src-dir=`pwd`/..

I see that the Python code should have changed the directory so I am not sure what went wrong. I will start debugging this tonight. We may also want to have our tests write output to a temporary directory rather than making any assumptions about current working directory.

[dianaclarke]

One of the tests failed because it makes an assumption that a target directory exists and it did not. My assumption is that somehow the current working directory was not correct. This is how I initiated the benchmark run.

~/git/arrow-rs/conbench$ conbench arrow-rs --src-dir=`pwd`/..

I see that the Python code should have changed the directory so I am not sure what went wrong. I will start debugging this tonight. We may also want to have our tests write output to a temporary directory rather than making any assumptions about current working directory.

If you pull the latest, I just removed the need for --src-dir. Hopefully that works for you.

$ cd ~/arrow-rs/conbench/
$ conda create -y -n conbench python=3.9
$ conda activate conbench
(conbench) $ pip install -r requirements.txt
(conbench) $ conbench arrow-rs

PS. I don't work on arrow at all anymore, so I better go back to my day job now – just didn't want to leave you hanging. Cheers!

[andygrove]

Thanks @dianaclarke I can take it from here for sure .. and it turns out this benchmark was failing even before this PR. I should have checked that first.

[dianaclarke]

Thanks @dianaclarke I can take it from here for sure .. and it turns out this benchmark was failing even before this PR. I should have checked that first.

Ah, I was getting an error with cargo bench in arrow-rs, but I assumed I just didn't have my env set up correctly (I don't know Rust aside from what I learned doing this).

But I can cargo bench in arrow-datafusion, so perhaps it's easier to start with this pull request:

• DataFusion + Conbench Integration datafusion#1791

[andygrove]

LGTM. After commenting out the failing csv_writer bench, conbench runs to completion and produces reports. A fix to csv_writer has already been merged to master.

[andygrove]

If there are no objections, I plan to merge this PR next week so that I can proceed to the next step of integrating this with CI. The changes in this PR are self-contained so I think it is low risk to merge.

[houqp]

Thank you @dianaclarke ! Great to see conbench finally lands in arrow-rs :)