Integrate Performance Tests in CI

[cdamus]

What it does

Adds a “Performance Tests” job to the build workflow in the project’s GitHub Actions to execute the browser and Electron start-up performance scripts, record and chart the results, and compare with past results for regression. The job is configured to

• Run only in the scheduled nightly build on the master branch
• Publish results in the GitHub pages at https://eclipse-theia.github.io/theia/tests/performance/
• On detection of measurements 150% longer than recent history, annotate the commit that was tested with a comment indicating that it may cause a performance regression
• The build does not fail on detection of regression. It is a warning condition only

Example output: an example of the performance history chart produced by these changes is available here.

A new dependency on the third-party github-action-benchmark GitHub action is added. It is configured to

• Commit and push performance history updates and visualization charts to the project’s GitHub pages. Note that this does not interfere with the /docs/next/ content already published on this branch. This can be seen here
• Annotate the HEAD commit with comments when performance regression is detected

Resolves #10443.

Contributed on behalf of STMicroelectronics.

How to test

Execute the performance scripts pretending the GitHub CI environment:

$ GITHUB_ACTIONS=true yarn performance:startup:browser && \
>   yarn performance:startup:electron

This will record performance measurements to the performance-result.json file that the GitHub benchmark action picks up for its history recording and analysis.

As the build job is configured not to run on PR branches but only on the master branch, testing the rest of the flow is more tricky. The behaviour can be observed in this fork in which the changes were initially drafted.

Review checklist

• As an author, I have thoroughly tested my changes and carefully followed the review guidelines

Reminder for reviewers

• As a reviewer, I agree to behave in accordance with the review guidelines

[cdamus]

Commit 29a5ef7 rebases the branch onto recent changes on master and resolves merge conflicts.

[Niklas-Dahlquist]

Bumping, much needed feature to be able to improve the startup performance, this is the first milestone to start to be able to track the duration to see what can be improved.

[cdamus]

Commit b41f101 rebases the original commit again following the release tag of 1.32.0 in the changelog, adding a new 1.33.0 section and putting the change entry there.

Commit 4cc485d then factors out the performance-test workflow into its own file with, for now, only the dispatch trigger. I've also made a few other updates to align with changes in the dependencies used by other builds (Node version, Ubuntu version, action versions) in the months since the original contribution.

[cdamus]

I don't recognize the test failures on Ubuntu as anything that I could have broken with this PR. Not sure what to do about that.

[vince-fugnitto]

The updates look good to me 👍
I'm fine with keeping it as a manual workflow dispatch for the time being.

[cdamus]

This time, one of the Linux test platforms that failed previously passed and one of the Mac test platforms that passed previously failed. Definitely seems unrelated to the changes in this PR.

[cdamus]

Commit 6531598 resolves the usual changelog conflict. And so I took the opportunity while to doing that to squash the branch. No other changes.