[C++] parquet-encoding-benchmark makes invalid accesses if repetitions > 1 #42140
Assignees
Milestone
Comments
pitrou
added a commit
to pitrou/arrow
that referenced
this issue
Jun 13, 2024
pitrou
added a commit
that referenced
this issue
Jun 13, 2024
…42141) ### Rationale for this change `parquet-encoding-benchmark` would make invalid memory accesses if more than one repetition per benchmark is requested by the user. This was initially noticed in #41205 : switching to jemalloc would crash the specific benchmark(s) on a macOS ARM machine. ### What changes are included in this PR? Make sure benchmark data initialization is idempotent. ### Are these changes tested? Locally using Valgrind. ### Are there any user-facing changes? No. * GitHub Issue: #42140 Authored-by: Antoine Pitrou <antoine@python.org> Signed-off-by: Antoine Pitrou <antoine@python.org>
|
Issue resolved by pull request 42141 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Describe the bug, including details regarding any error messages, version, and platform.
When looking at some continuous benchmarking failures on macOS ARM, I came to the conclusion that
parquet-encoding-benchmarkcan crash if more than one benchmark repetition is requested.I managed to reproduce the issue locally on x86 (Ubuntu 22.04) using this command line:
$ ARROW_DEFAULT_MEMORY_POOL=system valgrind <build dir>/relwithdebinfo/parquet-encoding-benchmark --benchmark_repetitions=6 --benchmark_min_time=0.050000s --benchmark_filter=BM_ArrowBinaryPlain.EncodeLowLevelComponent(s)
C++
The text was updated successfully, but these errors were encountered: