Polish better decoder (Testing)

[zhuqi-lucas]

Which issue does this PR close?

• Related to Parquet decoder / decoded page Cache #7363

Rationale for this change

What changes are included in this PR?

Are there any user-facing changes?

[alamb]

🤖 ./gh_compare_arrow.sh Benchmark Script Running
Linux aal-dev 6.11.0-1013-gcp #13~24.04.1-Ubuntu SMP Wed Apr 2 16:34:16 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux
Comparing polish_better_decoder (0c3aa9b) to cee5124 diff
BENCH_NAME=arrow_reader_row_filter
BENCH_COMMAND=cargo bench --all-features --bench arrow_reader_row_filter
BENCH_FILTER=
BENCH_BRANCH_NAME=polish_better_decoder
Results will be posted here when complete

[alamb]

🤖: Benchmark completed

Details

group                                                                                 main                                   polish_better_decoder
-----                                                                                 ----                                   ---------------------
arrow_reader_row_filter/Composite/all_columns/async                                   1.04      2.9±0.02ms        ? ?/sec    1.00      2.8±0.03ms        ? ?/sec
arrow_reader_row_filter/Composite/all_columns/sync                                    1.00      3.3±0.02ms        ? ?/sec    1.00      3.3±0.03ms        ? ?/sec
arrow_reader_row_filter/Composite/exclude_filter_column/async                         1.00      2.6±0.02ms        ? ?/sec    1.03      2.6±0.01ms        ? ?/sec
arrow_reader_row_filter/Composite/exclude_filter_column/sync                          1.00      2.8±0.02ms        ? ?/sec    1.00      2.8±0.01ms        ? ?/sec
arrow_reader_row_filter/ModeratelySelectiveClustered/all_columns/async                1.03      3.1±0.02ms        ? ?/sec    1.00      3.0±0.01ms        ? ?/sec
arrow_reader_row_filter/ModeratelySelectiveClustered/all_columns/sync                 1.00      3.3±0.02ms        ? ?/sec    1.01      3.3±0.02ms        ? ?/sec
arrow_reader_row_filter/ModeratelySelectiveClustered/exclude_filter_column/async      1.00      2.9±0.01ms        ? ?/sec    1.01      2.9±0.01ms        ? ?/sec
arrow_reader_row_filter/ModeratelySelectiveClustered/exclude_filter_column/sync       1.00      3.1±0.02ms        ? ?/sec    1.01      3.1±0.02ms        ? ?/sec
arrow_reader_row_filter/ModeratelySelectiveUnclustered/all_columns/async              1.00      6.5±0.04ms        ? ?/sec    1.00      6.5±0.03ms        ? ?/sec
arrow_reader_row_filter/ModeratelySelectiveUnclustered/all_columns/sync               1.00      6.5±0.02ms        ? ?/sec    1.03      6.7±0.06ms        ? ?/sec
arrow_reader_row_filter/ModeratelySelectiveUnclustered/exclude_filter_column/async    1.00      5.7±0.03ms        ? ?/sec    1.00      5.7±0.04ms        ? ?/sec
arrow_reader_row_filter/ModeratelySelectiveUnclustered/exclude_filter_column/sync     1.00      5.7±0.04ms        ? ?/sec    1.03      5.9±0.03ms        ? ?/sec
arrow_reader_row_filter/PointLookup/all_columns/async                                 1.00      2.1±0.01ms        ? ?/sec    1.00      2.1±0.01ms        ? ?/sec
arrow_reader_row_filter/PointLookup/all_columns/sync                                  1.00      2.2±0.01ms        ? ?/sec    1.01      2.2±0.01ms        ? ?/sec
arrow_reader_row_filter/PointLookup/exclude_filter_column/async                       1.00      2.0±0.01ms        ? ?/sec    1.00      2.0±0.01ms        ? ?/sec
arrow_reader_row_filter/PointLookup/exclude_filter_column/sync                        1.00      2.2±0.01ms        ? ?/sec    1.01      2.2±0.01ms        ? ?/sec
arrow_reader_row_filter/SelectiveUnclustered/all_columns/async                        1.02      3.2±0.03ms        ? ?/sec    1.00      3.1±0.03ms        ? ?/sec
arrow_reader_row_filter/SelectiveUnclustered/all_columns/sync                         1.00      3.4±0.04ms        ? ?/sec    1.01      3.4±0.02ms        ? ?/sec
arrow_reader_row_filter/SelectiveUnclustered/exclude_filter_column/async              1.00      2.8±0.01ms        ? ?/sec    1.03      2.9±0.02ms        ? ?/sec
arrow_reader_row_filter/SelectiveUnclustered/exclude_filter_column/sync               1.00      3.0±0.02ms        ? ?/sec    1.01      3.0±0.02ms        ? ?/sec
arrow_reader_row_filter/UnselectiveClustered/all_columns/async                        1.01      8.0±0.05ms        ? ?/sec    1.00      7.8±0.06ms        ? ?/sec
arrow_reader_row_filter/UnselectiveClustered/all_columns/sync                         1.01      8.3±0.07ms        ? ?/sec    1.00      8.1±0.07ms        ? ?/sec
arrow_reader_row_filter/UnselectiveClustered/exclude_filter_column/async              1.01      7.7±0.04ms        ? ?/sec    1.00      7.6±0.04ms        ? ?/sec
arrow_reader_row_filter/UnselectiveClustered/exclude_filter_column/sync               1.01      8.0±0.04ms        ? ?/sec    1.00      7.9±0.05ms        ? ?/sec
arrow_reader_row_filter/UnselectiveUnclustered/all_columns/async                      1.01      3.1±0.03ms        ? ?/sec    1.00      3.1±0.03ms        ? ?/sec
arrow_reader_row_filter/UnselectiveUnclustered/all_columns/sync                       1.00      3.4±0.01ms        ? ?/sec    1.02      3.5±0.03ms        ? ?/sec
arrow_reader_row_filter/UnselectiveUnclustered/exclude_filter_column/async            1.00      2.8±0.01ms        ? ?/sec    1.02      2.9±0.02ms        ? ?/sec
arrow_reader_row_filter/UnselectiveUnclustered/exclude_filter_column/sync             1.00      3.0±0.02ms        ? ?/sec    1.01      3.1±0.02ms        ? ?/sec
arrow_reader_row_filter/Utf8ViewNonEmpty/all_columns/async                            1.03     23.9±0.14ms        ? ?/sec    1.00     23.1±0.14ms        ? ?/sec
arrow_reader_row_filter/Utf8ViewNonEmpty/all_columns/sync                             1.00     23.9±0.13ms        ? ?/sec    1.00     23.8±0.12ms        ? ?/sec
arrow_reader_row_filter/Utf8ViewNonEmpty/exclude_filter_column/async                  1.00     15.2±0.06ms        ? ?/sec    1.00     15.2±0.11ms        ? ?/sec
arrow_reader_row_filter/Utf8ViewNonEmpty/exclude_filter_column/sync                   1.01     15.2±0.10ms        ? ?/sec    1.00     15.0±0.08ms        ? ?/sec

[alamb]

🤖: Benchmark completed

Interesting, the decoder cache doesn't seem to help much on my test machine (which is some crappy gcp VM). I couldn't reproduce the results listed on #7363 (comment) 🤔

[zhuqi-lucas]

🤖: Benchmark completed

Interesting, the decoder cache doesn't seem to help much on my test machine (which is some crappy gcp VM). I couldn't reproduce the results listed on #7363 (comment) 🤔

Thank you @alamb , it seems no obvious improvement compares to main. This branch only improve PointLookup for 1000000 line big data set comparing to original better-decode.

I agree, we need to find how to mock clickbench result from arrow-rs side.

[alamb]

I agree, we need to find how to mock clickbench result from arrow-rs side.

• Filed arrow_reader_row_filter benchmark doesn't capture page cache improvements #7460