Try to reduce Unit test CPU requests

📌 Unpack recipe test is slow Active needs to land first - that test is hogging all the CPU.

OK there's a definite penalty with 2 CPUs, but 4 CPUs seems pretty safe overall - most runs finishing in 35s, saw a couple take 1m.

One day we'll get at 📌 Autodetect available CPUs in run-tests.sh for concurrency Active and things will be simpler.

With 📌 Autodetect available CPUs in run-tests.sh for concurrency Active , unit tests would run in 27-29 seconds on 64 cores, https://git.drupalcode.org/issue/drupal-3526459/-/pipelines/534099

Potentially we could reach 20-25 secs if we could optimize two laggers

• Drupal\Tests\Core\Command\QuickStartTest
• Drupal\Tests\Core\Recipe\RecipeQuickStartTest

This issue will reduce cores to 32 if I am not mistaken - curious to see what it will mean then.

Potentially we could reach 20-25 secs if we could optimize two laggers

Drupal\Tests\Core\Command\QuickStartTest
Drupal\Tests\Core\Recipe\RecipeQuickStartTest

I wonder if it's worth discussing moving those to build tests, they seem build-test-ish to me even if they technically don't have to be. It would not only mean there's no slow test floor to the runtimes, but free up a CPU to run other tests, so would be a double saving during the lint stage. They'd be fast build tests instead of slow unit tests then.

The other lint stage saving we could make (CPU time, probably not wall time) is 📌 Use artifact caching for phpcs Active .

Opened 📌 Consider moving quickstart tests to build tests Active .

Committed and pushed 2f667837672 to 11.x and c02fefb64be to 11.2.x. Thanks!

Backported to 11.2.x to keep test stuff in sync.

Despite this reducing the number of CPUs, after 📌 Consider moving quickstart tests to build tests Active that moved the two tests to the Build suite, now 📌 Autodetect available CPUs in run-tests.sh for concurrency Active has a record execution time of 14 secs for core PHPUnit tests under PHP 8.5:

https://git.drupalcode.org/issue/drupal-3526459/-/jobs/5815987/viewer

8.3 and 8.4 are slightly below 30 secs, but I think it's the actual runner making the real difference here.