Joe McDonnell created IMPALA-15188:
--------------------------------------
Summary: report_benchmark_results.py should have an option to skip
HTML profile diff generation
Key: IMPALA-15188
URL: https://issues.apache.org/jira/browse/IMPALA-15188
Project: IMPALA
Issue Type: Improvement
Components: Infrastructure
Affects Versions: Impala 5.0.0
Reporter: Joe McDonnell
When running the perf-AB-test for TPC-DS, the report_benchmark_results.py
script takes a long time to produce the difference report. It turns out that
most of that time is in the save_runtime_diffs() function that is generating an
HTML profile difference. These are not preserved by the perf-AB-test job, so
this is useless for that circumstance. We should add an option to control this
output and skip it by default.
When running it locally, the output HTML is difficult to work with. Chrome
struggles even with local files on a fast machine. We should consider just
ripping this out entirely.
{noformat}
Running this on a TPC-DS run with 25 iterations, i.e.
$ time tests/benchmark/report_benchmark_results.py --input_result_file
~/7d59f4a382568ab4aaaa0af68101bc9f787bbd5c.json --reference_result_file
~/087c32bcf44f9a5af5df54ccfb4e9bc27ecd3ab1.json
Time with the save_runtime_diffs() output:
real 10m35.059s
user 10m22.647s
sys 0m11.676s
Time without the save_runtime_diffs() output:
real 0m19.441s
user 0m11.552s
sys 0m7.228s{noformat}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
