voonhous opened a new pull request, #19918:
URL: https://github.com/apache/hudi/pull/19918

   ### Describe the issue this Pull Request addresses
   
   Closes #19528. Part of the CI improvement epic #19524. Stacked on #19909; 
the commits before "ci: list the slowest tests of each job on the run page" 
belong to that stack and drop out on rebase.
   
   The three bullets of #19528, and where each lands:
   - rebalance the splits by measured job duration: #19892, underneath this PR 
in the stack (longest job 77 to 41 minutes);
   - report the slowest tests per job: this PR;
   - early signal for an obviously broken PR: measured below and not proposed, 
because the flattened graph already surfaces a compile failure about eight 
minutes into every job and a gate would cost a queue trip per run to save 
runner-minutes on about one run in two hundred.
   
   Today the per-class times behind a rebalance have to be pulled out of job 
logs by hand, and a job's long tail (a 617-second TestColumnStatsIndexWithSQL, 
a 341-second ITTestRepairsCommand) is invisible unless someone goes looking.
   
   **Example:**
   <img width="887" height="1033" alt="image" 
src="https://github.com/user-attachments/assets/cd7d0881-18f5-4a39-b758-590741dfed45";
 />
   
   
   ### Summary and Changelog
   
   - `scripts/ci/test_timing_summary.py`: walks every `target/surefire-reports` 
directory, which surefire and the scalatest-maven-plugin both fill with JUnit 
XML, and prints the slowest classes and the slowest individual tests as 
Markdown. With `GITHUB_STEP_SUMMARY` set it appends the same text there. Works 
locally after `mvn test` too.
   - Every job that runs tests, 14 in bot.yml and 3 in java_ci_engines.yml, 
gets one step after its test steps, `if: always()`, so the tables appear on the 
run's summary page for green and red jobs alike. The commented `[CI-TRIM]` 
lanes get the same step. The two bundle validation jobs run shell scripts, not 
surefire, and are left alone.
   
   ### Impact
   
   No change to what runs or how long it takes: the walk prunes below `target/` 
except `surefire-reports`, and finishes in well under a second on a built 
checkout. What changes is that the slow tail of every job is one click away on 
the run page, and the next rebalance reads its numbers from there instead of 
from log scraping.
   
   On the third bullet of #19528, early signal for broken PRs: across 186 
recent runs, one was broken at the build step (at least half its jobs failing 
to compile), and 50 of 2794 job executions failed at their build, almost all 
isolated download or resolution hiccups. A compile gate ahead of the test jobs 
would save runner-minutes on about one run in two hundred while adding a queue 
trip to every run, so it is not proposed here. With the flattened graph a 
compile failure already surfaces about eight minutes into every job.
   
   ### Risk Level
   
   low
   
   A reporting step that cannot fail the job: `if: always()` runs it after the 
tests, and a parse error on a report is skipped, not raised.
   
   ### Documentation Update
   
   none
   
   ### Contributor's checklist
   
   - [x] Read through [contributor's 
guide](https://hudi.apache.org/contribute/how-to-contribute)
   - [x] Enough context is provided in the sections above
   - [x] Adequate tests were added if applicable
   


-- 
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]

Reply via email to