HippoBaro opened a new pull request, #11449: URL: https://github.com/apache/arrow-rs/pull/11449
# Which issue does this PR close? - Spawn off from #11267 # Rationale for this change The Arrow writer benchmarks can show substantial performance drift based solely on execution order. On the benchmark host, running the full suite versus a filtered subset produced up to ~2x differences for the same benchmarks, depending on allocator state left by earlier cases. # What changes are included in this PR? Use jemalloc for the Arrow writer benchmark on Linux so its behavior can be configured explicitly. Use a single arena, disable background threads, and disable dirty and muzzy page decay to retain freed pages for reuse. With this configuration, the large execution-order-dependent variance was no longer reproducible on the benchmark host. Keep the dependency development-only and the global allocator override local to the benchmark executable; library consumers are unaffected. # Are these changes tested? All tests passing. # Are there any significant user-facing changes? None. -- 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]
