-
Notifications
You must be signed in to change notification settings - Fork 116
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Modernizing the project #264
Comments
FYI I'm trying the alternative approach on a personal repository with a local Jenkins instance. Proof of concept will be |
Hooray! Thanks for getting involved, sharing your initial pain and suggestions 👍 No doubt the github-bot age has started to become obvious, and lots has changed since its inception, e.g. GitHub Actions like you mention. And the pain of setting it up and even grasping how it works, hasn't been great for new comers, hence the low activity in the project. To be honest, I've questioned the actual value of the github-bot as is, when we now got GitHub Actions at our disposal. Not having an express server running on a server somewhere, would it in self be a success in my eyes. But I've mostly used Actions for trivial things like running tests and whatnot, so its shortcomings that you're referring, isn't something I'm familiar with yet. Worth mentioning that I'm not opinionated about keeping the github-bot as is, refactor or even replaced. As long as we find an approach that serves the collaborator community what it needs, I'm all in. In short I'm +1 to all your three concrete suggestions. Regarding your point about deprecated dependencies and callback vs Your suggestion of transitioning into an Action relay sounds really interesting and isn't something that has struck my mind! I like the idea of moving stuff into the projects themselfs, instead of centralising it into this github-bot project. From a glance it sounds like something that could ease the pain of understanding how things work, tho maybe a bit harder to test properly? Having automated tests verifying things still works changes are pushed, has been one of the pros by having the project as is IMO. What's your gut feeling and favourite amongst the suggestions you're proposing? |
My gut feeling is that if we can make everything work with actions, and we don't start hitting Actions limits all the time, it should be easier to maintain and it would be easier for folks to contribute. Testing is a good point, although some of the features (labeling for example) are not crucial enough IMO for tests to outweigh the lower contributors poll. Actions are testable too, so we could test the ones that are more critical to our workflow. |
Fair point. Also sure we'd find a creative way to perform some kind of automated tests if we really need to, meaning things blow up more often than we'd like too.. As said in #258, those changes are planned to land ASAP unless anyone objects. But say we want to go in the Actions relay direction long term, and you don't find any obvious showstoppers in the proof of concept you're working on, will you let us know when others can join the fun and contribute? Would be cool to hear your thoughts on the way forward; new endpoint in the existing github-bot project (followed by deleting lots of code) or a new project altogether etc? |
As soon as I get the proof of concept working I'll share it on |
Cool, looking forward to seeing it in practise! |
In our quest of delegating more to the actual repositories w/GitHub Actions and experimenting with making the bot an Actions relay, it has also dawned on me that our current infrastructure situation feels overkill and clunky. Although running it on the Node.js org infrastructure works, there's definitively a bus factor involved and it's not trivial to fully grasp. Maintenance is also time consuming and rarely prioritised. Except for the attempt-backport feature, that I've suggested be removed in #288, there's little reason for the github-bot to have a dedicated host and in practise be a full fledged webserver. From a distance it feels like something more nimble like Netlify functions and the like, would have made quite a difference and significantly reduced the bus factor. As always, the grass always feels greener on the other side, but I'm having trouble seeing the obvious downsides to something like that compared to what we've got at the moment. |
As long as the bot doesn't have access to security releases, I think that's fine? It's something that should be brought up to nodejs/build, if we move it should be to a Node.js Netlify (or w/e service we decide to use) account. |
Myeah, sounds like a good idea 👍 |
First of, I love our GitHub bot. It's an essential piece of our infrastructure, and it can fill the gaps where GitHub Actions don't work well (the whole read-only, no secrets situation with forked PRs). With that being said, I have been trying to contribute to the project and the entry barrier seems somewhat high:
github
,request
, there might be others too)github
is not only deprecated, it lacks documentation. To find out which method to use, developers need to look into the TypeScript definition and then figure out which GitHub API is being called to read the documentation on https://developer.github.com/v3/nodejs
as org andnode
as repository, which makes it impossible for users to test without finding these conditionals and changing them/commenting those outThe project is also heavily callback-oriented, some parts would probably benefit from changing to an
async/await
-oriented implementation (I know this is more controversial, so I won't push too much on it).Based on the points above, I have some suggestions to modernize the project:
nodejs/node-auto-test
. Anyone who wants to collaborate would connect to that relay, and they would receive the appropriate tokens needed for testing. We could add checks so that only certain teams are allowed to request access. We can also have a separate user with permissions limited tonodejs/node-auto-test
, to prevent folks from mistakenly affecting other repositories. The same relay could be used for Jenkins hooks.nodejs/node
, if we moved the labels definition tonodejs/node
(so that the bot loads the definition when needed), it would be easier for collaborators to include/remove definitions for new files, and it would also allow the bot to detect if that repository supports labeling or not.async/await
API might be harder. We could either start a new branch from scratch, or try to modernize different pieces individually. Either way it will need some coordination efforts as well as a lot of work to get it working properly.As an alternative, we could turn the
github-bot
into an Actions relay: it would receive events from GitHub and Jenkins, and would forward those events to the repository dispatch API. This way, we could define everything as Actions on the respective repositories, circumventing the gaps with Actions on forked PRs and being able to define Actions for Jenkins events.What do folks think? cc @nodejs/github-bot
The text was updated successfully, but these errors were encountered: