GH-35636: [C++] Extract two expensive test suites from compute-vector-test

[felipecrv]

Rationale for this change

arrow-compute-vector-test is too big and takes a long time to run because of that.

What changes are included in this PR?

Extracting two tests.

Timings on my machine (Debug builds with ASAN).

debug/arrow-compute-vector-test > /dev/null  11.54s user 0.47s system 99% cpu 12.023 total
debug/arrow-compute-vector-sort-test > /dev/null  13.30s user 0.26s system 99% cpu 13.579 total
debug/arrow-compute-vector-selection-test > /dev/null  6.97s user 0.22s system 99% cpu 7.207 total

Are these changes tested?

Yes.

Are there any user-facing changes?

No.

• Closes: [CI][C++] Nightly test-conda-cpp-valgrind is flaky due to arrow-compute-vector-test taking closer to 500 seconds #35636

[pitrou]

@github-actions crossbow submit -g cpp

[pitrou]

At some point we might want to create an object library for test_util.cc instead of recompiling it for each target. Can be left as a subsequent PR.

[felipecrv]

I was tempted to do it, but I was afraid it would make things worse because test_util.cc is so small.

[pitrou]

That's possible. We can revisit later :-)

[github-actions]

Revision: 021fb90

Submitted crossbow builds: ursacomputing/crossbow @ actions-ccd2ce168e

Task	Status
test-alpine-linux-cpp
test-build-cpp-fuzz
test-conda-cpp
test-conda-cpp-valgrind
test-cuda-cpp
test-debian-11-cpp-amd64
test-debian-11-cpp-i386
test-fedora-35-cpp
test-ubuntu-20.04-cpp
test-ubuntu-20.04-cpp-20
test-ubuntu-20.04-cpp-bundled
test-ubuntu-20.04-cpp-minimal-with-formats
test-ubuntu-20.04-cpp-thread-sanitizer
test-ubuntu-22.04-cpp

[pitrou]

I think you need to run cmake-format for the lint step to pass... or apply the following patch:

diff --git a/cpp/src/arrow/compute/kernels/CMakeLists.txt b/cpp/src/arrow/compute/kernels/CMakeLists.txt
index a84b36a52..9084f279a 100644
--- a/cpp/src/arrow/compute/kernels/CMakeLists.txt
+++ b/cpp/src/arrow/compute/kernels/CMakeLists.txt
@@ -76,14 +76,9 @@ add_arrow_compute_test(vector_test
                        select_k_test.cc
                        test_util.cc)
 
-add_arrow_compute_test(vector_sort_test
-                       SOURCES
-                       vector_sort_test.cc
-                       test_util.cc)
+add_arrow_compute_test(vector_sort_test SOURCES vector_sort_test.cc test_util.cc)
 
-add_arrow_compute_test(vector_selection_test
-                       SOURCES
-                       vector_selection_test.cc
+add_arrow_compute_test(vector_selection_test SOURCES vector_selection_test.cc
                        test_util.cc)
 
 add_arrow_benchmark(vector_hash_benchmark PREFIX "arrow-compute")

[pitrou]

Nice timings here too:

/build/build-test ~/arrow/dev/cpp
Test project /build/build-test
    Start 26: arrow-compute-vector-sort-test
    Start 25: arrow-compute-vector-test
    Start 27: arrow-compute-vector-selection-test
1/3 Test #27: arrow-compute-vector-selection-test ...   Passed    2.94 sec
2/3 Test #25: arrow-compute-vector-test .............   Passed    4.84 sec
3/3 Test #26: arrow-compute-vector-sort-test ........   Passed    6.15 sec

100% tests passed, 0 tests failed out of 3

Label Time Summary:
arrow_compute    =  13.94 sec*proc (3 tests)
unittest         =  13.94 sec*proc (3 tests)

Total Test time (real) =   6.15 sec
~/arrow/dev/cpp

real	0m6,166s
user	0m13,900s
sys	0m0,092s

[felipecrv]

I think you need to run cmake-format for the lint step to pass... or apply the following patch:

I ran the CMake linter locally and was so confused: it only says the checks doesn't pass.

[pitrou]

The CI failures seem unrelated, I'll merge. Thanks @felipecrv !

[conbench-apache-arrow]

Conbench analyzed the 6 benchmark runs on commit 1bfa241b.

There were 9 benchmark results indicating a performance regression:

• Commit Run on ursa-i9-9960x at 2023-07-07 05:54:58Z

  • engine=arrow, format=native, language=R, memory_map=False, query_id=TPCH-16, scale_factor=1

• Commit Run on arm64-m6g-linux-compute at 2023-07-03 19:09:54Z

  • params=simple_expression/batch_size:10000/null_prob:0/bool_true_prob:50/real_time, source=cpp-micro, suite=arrow-acero-filter-benchmark

• and 7 more (see the report linked below)

The full Conbench report has more details.