Come together with the global Drupal community in Rotterdam, 28 Sept – 1 Oct 2026. Sessions, contribution, connection, and Early Bird savings until 8 June.Problem/Motivation
In a simple functional test run there are 11 container rebuilds. Let's try to reduce that number.
Steps to reproduce
Proposed resolution
Remaining tasks
User interface changes
Introduced terminology
API changes
Data model changes
Release notes snippet
Issue fork drupal-3571994
Show commands
Start within a Git clone of the project using the version control instructions.
Or, if you do not have SSH keys set up on git.drupalcode.org:
changes
Comments
Comment #3
longwaveThis change reduces the number of container builds from 11 to 9.
Comment #4
longwaveNot very scientific yet, but looks promising:
Functional tests on main:
https://git.drupalcode.org/project/drupal/-/pipelines/734148/builds
Minimum 5:20
Maximum 6:17
Functional tests on this MR:
https://git.drupalcode.org/issue/drupal-3571994/-/pipelines/735093/builds
Minimum 3:11
Maximum 4:48
Comment #5
longwaveTried some further combining of container rebuilds, but in more complicated test setups this isn't going to work. Just avoiding rebuilding the InstallerKernel twice is good enough for now.
Comment #6
longwaveThe latest changes get it down from 11 to 8 builds in a trivial test.
Measuring the job runtime isn't very useful because of variance in container startup time. If instead we look at the actual runtimes reported in the logs:
Functional tests on main:
https://git.drupalcode.org/project/drupal/-/pipelines/734148/builds
BrowserTestBase: 3:34 to 4:31
WebDriverTestBase: 3:37 to 4:15
Functional tests on this branch:
https://git.drupalcode.org/issue/drupal-3571994/-/pipelines/735196/builds
BrowserTestBase: 1:47 (!) to 4:47
WebDriverTestBase: 3:25 to 4:42
Unfortunately it appears to depend more on hardware availability and time of day than anything else, maybe this is still worth it though?
Comment #7
catchYeah there seem to be huge differences depending on time of day and hardware, but if the floor goes down, that is a good sign it's doing what it's supposed to.
Can't see anything to complain about, so moving to RTBC.
We have a noticeable regression in functional test times between main and 11.x (unless that is also unlucky hardware variation, could not detect anything locally), but looks like this is equally applicable to main and 11.x so it'll just mean a different baseline for both branches.
Comment #8
alexpottCommitted and pushed c84deaef44f to main and 70b8759d613 to 11.x. Thanks!