[
https://issues.apache.org/jira/browse/ARROW-15785?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17498146#comment-17498146
]
Jonathan Keane edited comment on ARROW-15785 at 2/25/22, 2:16 PM:
------------------------------------------------------------------
I think this is the PR that introduced the regression (though I might be
totally off or it's a different regression...)
https://github.com/apache/arrow/pull/11991#issuecomment-1009216946
And the conbench run:
https://conbench.ursa.dev/compare/runs/c4d5e65d088243259e5198f4c0e219c9...5a1c693586c74471b7c8ba775005db54/
We should probably have the conbench bot alert more loudly that there are
regressions of this magnitude. That 5% there is supposed to indicate that
there's an issue, but we might have that set too low such that there's alarm
fatigue or|and we should alert louder when there are this many high-change
benchmarks (e.g. the file-read benchmark z-scores range from a -76 to -759, and
we alert at -5)
was (Author: jonkeane):
I think this is the PR that introduced the regression (though I might be
totally off or it's a different one...)
https://github.com/apache/arrow/pull/11991#issuecomment-1009216946
And the conbench run:
https://conbench.ursa.dev/compare/runs/c4d5e65d088243259e5198f4c0e219c9...5a1c693586c74471b7c8ba775005db54/
We should probably have the conbench bot alert more loudly that there are
regressions of this magnitude. That 5% there is supposed to indicate that
there's an issue, but we might have that set too low such that there's alarm
fatigue or|and we should alert louder when there are this many high-change
benchmarks (e.g. the file-read benchmark z-scores range from a -76 to -759, and
we alert at -5)
> [Benchmarks] Add conbench benchmark for single-file parquet reads
> -----------------------------------------------------------------
>
> Key: ARROW-15785
> URL: https://issues.apache.org/jira/browse/ARROW-15785
> Project: Apache Arrow
> Issue Type: Improvement
> Components: Benchmarking
> Reporter: Weston Pace
> Assignee: Weston Pace
> Priority: Major
>
> Release 7.0.0 introduced a regression in parquet single file reads. We
> should add a macro-level benchmark that does single-file reads to help us
> detect this in the future.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
