[DRAFT] [RFD]: Does Doodba as a base still make sense for us?

[theangryangel]

This is a draft tracking issue/note-taking only at this point. It will be adjusted into an appropriate format.

---

• Doodba serves us very well, it generally just works, but it is:

  • Different vs the standard Odoo container, which has confused other partners. Can we get it closer?
  • Does a lot of clever stuff, some of which we don't need
  • In the very very rare occasion when there's a problem, we need to work-around them
  • Does a lot more in python when there are other tools, like envsubst, run-parts, etc. that perhaps might be better suited
  • Some Glo team members struggle with some of the complexities
  • There's stuff in the base image we don't care about, can we shrink it?

• We spend a lot of CI minutes on tests, etc. can we speed things up?

  • Can we speed up build at all?
  • Can we speed up tests?

• Pulling images on Azure can be quite slow sometimes due to their size. It seems inconsistent and probably hub.docker.com to blame. Can we shrink things?

Unorganised thoughts:

• Rather than having any base, we could just template the Dockerfile into the project? Is that worth it? Probably not, some stuff should stay in an onbuild, but worth testing.
• Can/should we use specific
• Should/could we use run-parts for entrypoint.d
  • We should avoid pg_isready to avoid "double ready" problems
• Use envsubst for config file generation from env vars to more easily add more support
• Can/should we git-aggregate and addons linking outside of container?
  • This would allow us to use https://github.com/acsone/git-autoshare more easily
• Can/should we use uv instead of pip for speed?
• Can/should we bring the layout/structure closer to a "standard" Odoo container
• Would a devcontainer be useful?
• Is there any benefit for us using jemalloc?
• Supporting OCA/queue job runner cli? What does that mean/look like?

[theangryangel]

To link into this, a lack of a pyproject.toml is becoming a frustration as we've started to litter the scaffolding and more and more tools are aligning towards pyproject.toml.

We've got an active experiment running in glodouk/odoo-addons where we do not use Odoo. It's semi-successful.

For odoo/odoo you need to jump through some hoops. Or use a patched version. For odoo/enterprise it's the same.

Odoo SA are showing near zero outward signs on improving this.

There is work from Stéphane Bidoul which helps. But again, see the above issues regarding Odoo SA.