Weekly build triage - Automation

[adamfarley]

Summary

To automate the process of build failure triage.

Note, this does include propagated red test framework failures, but will not include unit test failure.

MVP

Steps 1-3, as per the full plan below.

Full Plan

• Step 1) A GitHub Action calls a script on Mondays.
• Step 2) The script identifies the latest automated build pipelines, limited to the last 7 days, for each current JDK major version.
• Step 3) The script then identifies which subjobs failed.
• Step 4) We then access each failed subjob and use a list of easily-modified patterns to identify useful output relevant to the failure.
  • 3-4ph: Draw up list of initial regexs and any (rough) relevant data.
  • 1-4ph: Test regexs and fix bugs.
  • 1pd: Approximate Total.
  • Actual time: 1.6pd
    • Note: Creating the logic to actually apply the regexs took longer than expected. Make sure to set aside time for supporting logic in future estimations.
• Step 5) Each pattern should be associated with metadata, including problem type, advice, any relevant issue links, and whether it is generally considered a preventable issue.
  • 2-3ph: Tidy metadata and identify preventables.
  • 1-2ph: Test metadata displays correctly.
  • 0.5pd: Approximate Total.
  • Actual time: 0.3pd
• Step 6) A summary is formatted into a table like the one at the top here.
  • 2-4ph: Write initial code to generate md table.
  • 1-3ph: Fix bugs.
  • 1ph: Output table to file.
  • 1-4ph: Test output works and is accurate.
  • 0.5-2pd: Approximate Total.
  • Actual time: 0.2pd
    • Note: The table doesn't seem as useful as a simple number, which is what the community asked for upstream.
• Step 7) All of the information (the link to the job, the useful output, and the metadata) then gets put after the table, in chunks.
  • 1-3ph: Write code and test. This should be simple.
  • Actual time: 1.5ph: It was, barring a single bug that's been fixed.
• Step 8) The summary and the full info list are then output from the script, where the GitHub Action collects them.
• Step 9) The action finally creates a build triage issue in the same repository, with the script output as the issue's contents.
• Step 10) Cake.
  • Cancelled this for diet's sake. 😞

[adamfarley]

Step 2 and 3 are complete and being tested. Code here: https://github.com/adamfarley/temurin-build/tree/build_triage_automation

Will likely spend the rest of the day adding regexs and plugging all this into an issue-generating github action.

[adamfarley]

Steps 1, 8, and 9 were tricky, but are now complete.

A PR for this issue's MVP commit can be found here.

For future improvements, I will be including a link to the branch used for said improvements. The commit messages on these branches can be used to gauge incremental status.

[adamfarley]

A minimal regex list, complete with metadata and logic, will be created here.

This fulfils steps 4 and 5.

[adamfarley]

Ok, the script is finished. I've created the documentation, updated the issue description field with the actual time it took (to refine estimates in the future), and created a PR to merge the changes: #3576

Further improvements to the build triage script will happen as time allows, and proposals/plans/results will be tracked in a new issue. (Link).