After merging your PR, Conbench analyzed the 4 benchmarking runs that have been run so far on merge-commit 7f5dd71.

Benchmarks with performance regressions

There was 1 possible performance regression, according to the lookback z-score method.

Benchmarks with regressions:

• Commit Run on amd64-m5-4xlarge-linux at 2025-07-11 03:04:38Z
  • partitioned-dataset-filter (R) with dataset=dataset-taxi-parquet, language=R, query=vignette

All benchmark runs analyzed:

• Commit Run on amd64-c6a-4xlarge-linux at 2025-07-11 02:52:03Z
• Commit Run on arm64-t4g-2xlarge-linux at 2025-07-11 03:10:34Z
• Commit Run on amd64-m5-4xlarge-linux at 2025-07-11 03:04:38Z
• Commit Run on test-mac-arm at 2025-07-11 03:10:50Z

Unstable benchmarks with performance regressions

The following benchmark results indicate a possible performance regression, but are known to sometimes produce false positives when applying the lookback z-score analysis.

• Commit Run on arm64-t4g-2xlarge-linux at 2025-07-11 03:10:34Z

  • BM_DeltaBitPackingDecode_Int64_Fixed (C++) with params=4096, source=cpp-micro, suite=parquet-encoding-benchmark
  • BM_ByteStreamSplitEncode_Float_Scalar (C++) with params=1024, source=cpp-micro, suite=parquet-encoding-benchmark
  • ExecuteScalarExpressionOverhead (C++) with params=complex_expression/rows_per_batch:100000/real_time/threads:1, source=cpp-micro, suite=arrow-acero-expression-benchmark
  • BM_PlainEncodingSpacedBoolean (C++) with params=32768/1000, source=cpp-micro, suite=parquet-encoding-benchmark
  • BM_ByteStreamSplitEncode_FLBA_Generic (C++) with params=<7>/1024, source=cpp-micro, suite=parquet-encoding-benchmark

• Commit Run on amd64-m5-4xlarge-linux at 2025-07-11 03:04:38Z

  • AllocateTouchDeallocate (C++) with params=/size:1048576/real_time/threads:8, source=cpp-micro, suite=arrow-memory-pool-benchmark