feat: Create separate repository for our in-house GitHub actions

[courtneypacheco]

This doc describes how we can reuse our in-house actions as our CI processes continue to grow.

[nathan-weinberg]

@courtneypacheco can you socialize this on Slack/Discord/the mailing lists/wherever you want so we can get community input?

[jjasghar]

This seems like a reasonable expectation, but we should probably socialize this more before adoption.

If anything, we should talk about it at the Tuesday Community meeting.

[markstur]

ci-actions repo sounds like a good idea to me

[nathan-weinberg]

+1 on this idea - I would vote on not publishing to GitHub Actions Marketplace - far prefer a single repo for the InstructLab org

[bbrowning]

I'm +1 generally to having a central place for all of our in-house GitHub actions. My only question is do we want this to be a repository just for GitHub actions, or a repository for centralized CI more generally which may initially only include some common GitHub actions but that may expand to be other things? Or, is the thought that we can encapsulate the majority of what could be common across CI setups within GitHub actions?

[ktdreyer]

Proposal looks great. It will be great to reduce duplication like in that free-disk-space example.

The more we keep the focus on GitHub Actions content, the easier it will be to lint etc and guide new contributors.

@bbrowning what other types of things were you picturing hosting in this repository?

[bbrowning]

@ktdreyer The other obvious thing we share across repos, besides GitHub actions themselves, are the workflow files. So, if we wanted to reduce duplication across the repositories by creating a couple of reusable workflows, would we also store those in this same place? Or would that be in a different centralized place?

The only reason I bring it up is that this is targeted specifically for our GitHub actions, so I just want to clarify if we mean only GitHub actions or if we mean this to house any shared CI bits, which may start out as only GitHub actions but grow in scope as necessary.

[courtneypacheco]

@bbrowning That makes sense.

If we do want to have a more generic repo for GitHub CI resources, we could certainly name this repo something like shared-ci-resources. In specific, if have lots of different shared CI resources (beyond the shared workflows and actions), then it could certainly get messy if we have a dedicated repo for each "category" of CI resource, so I can see why a shared-ci-resources repo might be helpful.

On the other hand, if we want to create releases for our shared workflows and actions, I would suggest creating separate repos in that case. Having releases (versioning) can reduce inconsistencies in behavior across repos. For example, if I update one of our in-house GH actions, that update may work for only 4 of the 5 repos it's used in, and break the 5th repo. Then the 5th repo has a non-working CI.

[danmcp]

The work being done here is another good example of what could be shared:

instructlab/training#419