DevOps: Add devcontainer

[ltalirz]

The devcontainer specification [1] is about making it easier for developers to get a first development environment up and running.

This gives (prospective) AiiDA developers and contributors a fully-fledged AiiDA development environment in two ways:

• Zero-install with Codespaces in the web browser.
  On GitHub, a user can simply click on "Code" => "Open with Codespaces", wait 5 minutes
• One-command cross-platform with VS Code + Docker.
  In the command palette, type "Open Folder in Container" (automatically pops up when opening folder in VS Code)

[1] https://containers.dev

[ltalirz]

a short video of how it works https://www.youtube.com/watch?v=ibkFSNLQsww&ab_channel=LeopoldTalirz

you can try it yourself at https://github.com/ltalirz/aiida-core/tree/feat/devcontainer-compose

Note: I first built a version with a single container in https://github.com/ltalirz/aiida-core/tree/feat/devcontainer but then abandoned it since the approach of the /sbin/my_init script does not work well together with the way VS Code treats the container.-

[ltalirz]

the duplication of the configure-aiida.sh can be removed before merging this;
it is only necessary because I'm using the aiida-core:main image as a base for the aiida service to speed up build time.

[ltalirz]

@yakutovicha I understand that the aiida-core Docker container is no longer used in the AiiDA lab, is that correct?

If that is the case, then we could probably at some point get rid of the aiida-prerequisites container in favor of a docker-compose.yml like the one I'm adding here that uses separate containers for postgres & rabbitmq.
The Dockerfile in the aiida-core repo could become self-contained and simply derive from something like python:3.9.

[yakutovicha]

@yakutovicha I understand that the aiida-core Docker container is no longer used in the AiiDA lab, is that correct?

That is correct 👍

[ltalirz]

@unkcpz @chrisjsewell @yakutovicha any thoughts on this PR?

It anyhow reuses the current aiida-core container, so the decision of deprecating aiida-prerequisites or not can be decoupled.

[chrisjsewell]

Looks fine to me, I've always championed multiple containers (via docker-compose) and VS Code

[unkcpz]

The Dockerfile in the aiida-core repo could become self-contained and simply derive from something like python:3.9.

I fully agree with this. I think instead of having the aiida image in aiidalab as I tried in aiidalab/aiidalab-docker-stack#319, it makes more sense to just have the parallel one base on the simple python base image for aiida. This may increase a bit duplication between aiidalab-docker-stack and the image in aiida-core, but that will make the images have clear usage for users and can be found from the corresponding repository.

[unkcpz]

I think the image aiidalab/base from https://hub.docker.com/r/aiidalab/base/tags is what we can put here in the future if we manage to move it from aiidalab to here, correct?

[ltalirz]

Almost. It would probably rather become FROM image in the .devcontainer/Dockerfile, since the devcontainer focuses on development (which requires e.g. some further dependencies installed).

[ltalirz]

Thanks for the feedback @unkcpz !

Given this, I will prepare this PR for review with minimal changes;
in a follow-up PR we can move from the current Dockerfile in the root of the repo to a more minimal one like the one you drafted in aiidalab/aiidalab-docker-stack#319

[ltalirz]

PR is now ready for review (the UI doesn't let me convert it from a draft)

[unkcpz]

thanks @ltalirz, I just have one question. I'll test locally it later.

[unkcpz]

Do this need to be kept here? Since it is useful for having data stored persistent I suggest keeping it and adding more details.

[ltalirz]

Yes, that was the intention. Will add further comments

[ltalirz]

in the end I decided to remove them as they are untested.

it's easy enough to reintroduce them and go to a setup with persistent data

[unkcpz]

Thanks, but it is a bit confusing for me how this codespace is working, if I understand correctly, the container is opened in a remote machine provided by GitHub. How is this persistent volumes mapping can mapping to the local DB file or local file repository?

[ltalirz]

Indeed this would have been intended for running the devcontainer locally through VS Code.

I am not aware of a way of persisting files after github codespaces deletes the containers.

[unkcpz]

@ltalirz Thanks for the update. I give it a test and it is super cool. Just one follow-up question on the volume mapping of the container.