- Issue created by @cmlara
- πΊπΈUnited States moshe weitzman Boston, MA
As an aside, all these same issues apply to gitlab_templates too. There we want our cake and eat it too. We want sites to automatically pickup changes to the templates but occasionally we break stuff, either by mistake or on purpose.
- πΊπΈUnited States cmlara
Opened #3404175: Adopt (semver) versioning for gitlab_templates β to discuss the gitlab_templates equivalent of this.
- πΊπΈUnited States cmlara
I will also note that if at the same time we move everything into a central
drupalci/php project we can avoid the issues of needing to know what base image type being used as drupalci/php:8.4-Ubuntu could be aliased to drupalci/php:8.4 which would make CI management significantly easier for version selection. - π«π·France andypost
IIRC the reason to keep only latest image is security because distros getting updates and PHP has monthly release cycle, so no reason to bloat storage...
So semver should be local and applicable for nightly somehow
- πΊπΈUnited States cmlara
IIRC the reason to keep only latest image is security because distros getting updates
Although the image is avaliable doesnβt mean a site has to use it. We see this all the time on DockerHub older releases are kept so they can be referred to in the future.
This is analogous to arguing all the Drupal 9 core releases should be deleted due to the fact they are known vulnerable.
and PHP has monthly release cycle, so no reason to bloat storage...
First off, retention and semver are different.
Semver would be about ensuring images are generally the same and compatible between build, PHP majors and minors. This could be done even with the current system of never retaining anything except the latest build.
Additional one could set a retention window for images (6 months of same βmajorβ and at least the last major of a semver for a few years?) to reduce storage bloat and yet still allow retaining images.
I will also add that shared storage of being in the same repository could allow image layer reuse further reducing size (this would require streamlining the builds a bit).