GH-45860: [C++] Respect CPU affinity in cpu_count and ThreadPool default capacity

[AntoinePrv]

Rationale for this change

We want the ThreadPool default capacity to follow the CPU affinity set by the user, if any.
For example:

$ python -c "import pyarrow as pa; print(pa.cpu_count())"
24
$ taskset -c 5,6,7 python -c "import pyarrow as pa; print(pa.cpu_count())"
3

What changes are included in this PR?

• Implement and expose CPU affinity detection as a utility function in arrow/io_util.h; on non-Linux platform, it returns Status::NotImplemented
• Use CPU affinity count, if available, to choose the default ThreadPool capacity

(note: based on original changes by Zihan Qi in PR #46034)

Are these changes tested?

By unit tests on CI, and by hand locally.

Are there any user-facing changes?

ThreadPool capacity now follows CPU affinity settings on Linux.

• GitHub Issue: [C++] Default thread count should count process affinity into account #45860

[AntoinePrv]

Moved from thread_pool.cc to standardize the detection of number of CPUs.

[AntoinePrv]

The check was and is unused.

[pitrou]

Just a quick comment here (I'm about to leave on holiday): given that we would like to get rid of CpuInfo, we should put this functionality (affinity detection and concurrency detection) in util/io_util.h instead.

[AntoinePrv]

Ha I understood from the previous discussion that this would remain the interface wrapping whichever underlying library.

[pitrou]

It might indeed, but I'm not sure anything is settled in that regard. Affinity is also not really a CPU property.

[pitrou]

Sorry for the delay @AntoinePrv , as I was on vacation.

Changes look good in principle, you'll find a set of minor comments and suggestions below.

[pitrou]

I have checked locally that it works (Ubuntu 24.04):

$ python -c "import pyarrow as pa; print(pa.cpu_count())"
24
$ taskset -c 5,6,7 python -c "import pyarrow as pa; print(pa.cpu_count())"
3
$ OMP_NUM_THREADS=9 taskset -c 5,6,7 python -c "import pyarrow as pa; print(pa.cpu_count())"
9
$ OMP_THREAD_LIMIT=9 taskset -c 5,6,7 python -c "import pyarrow as pa; print(pa.cpu_count())"
3
$ OMP_THREAD_LIMIT=2 taskset -c 5,6,7 python -c "import pyarrow as pa; print(pa.cpu_count())"
2

[pitrou]

@AntoinePrv Please don't mention usernames in the PR description, as they will end up in the final commit message and trigger annoying notifications later on.

[pitrou]

+1, thanks for the update @AntoinePrv !

[pitrou]

CI failure is unrelated.

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 4 benchmarking runs that have been run so far on merge-commit 8509ca4.

There weren't enough matching historic benchmark results to make a call on whether there were regressions.

The full Conbench report has more details.