- Universal Blue

Universal Blue•9mo ago

j0rge

j0rgeOP•4/18/25, 9:33 PM

This is awesome!!!!!!

j0rgeOP•4/18/25, 9:33 PM

automatic

j0rgeOP•4/18/25, 9:34 PM

whenever @Carl George gets us a new image, we kick off a build.

Jj0rge Click to see attachment

tulip•4/18/25, 9:42 PM

i think they do that on centos-bootc :)

tulip•4/18/25, 9:42 PM

thats really nice

j0rgeOP•4/18/25, 9:42 PM

Like, that build is just automated now

j0rgeOP•4/18/25, 9:43 PM

man dude if we could go cron-less across the board and just build when the stuff actually needs to be built

j0rgeOP•4/18/25, 9:46 PM

How many hours have we spent after fixing something in akmods to then babysit the build, so we can click main to rebuild, and then click the downstream images. Hours.

Jj0rge man dude if we could go cron-less across the board and just build when the stuff...

tulip•4/18/25, 9:54 PM

YESYESDYES

Jj0rge How many hours have we spent after fixing something in akmods to then babysit th...

Robert•4/18/25, 9:56 PM

https://github.com/ublue-os/main/pull/792
Starting on this

Won't be ready to merge today, but making a start

GitHub

chore: scaffold image-versions.yaml file by p5 · Pull Request #792...

Thank you for contributing to the Universal Blue project!
Please read the Contributor's Guide before submitting a pull request.

j0rgeOP•4/18/25, 9:56 PM

ok even the old ones huh?

Robert•4/18/25, 9:56 PM

The YAML isn't the cleanest, but it's not meant to be maintained by people

j0rgeOP•4/18/25, 9:57 PM

right

Jj0rge ok even the old ones huh?

Robert•4/18/25, 9:57 PM

Might as well. It's easier to roll it out across the board than build in the logic to exclude older images

j0rgeOP•4/18/25, 9:57 PM

I met the creator of yaml in london

j0rgeOP•4/18/25, 9:58 PM

ok so question

j0rgeOP•4/18/25, 9:58 PM

let's say you fix something on akmods

j0rgeOP•4/18/25, 9:58 PM

but some of the builders go red or we get a dud that takes a while.

j0rgeOP•4/18/25, 9:59 PM

is there a chance that the downstream bot could accidentally do it twice.

j0rgeOP•4/18/25, 10:00 PM

like main rebuilds but then the matrix finishes like 10m later or something then would it try to rebuild all of main again?

Robert•4/18/25, 10:00 PM

You mean the merge queue builds and pushes an image, but other images fail so the merge queue is cancelled and reran?

j0rgeOP•4/18/25, 10:00 PM

ok so let's say silverblue, kinoite, and base-main for three examples.

j0rgeOP•4/18/25, 10:00 PM

bot fires off, automerges

j0rgeOP•4/18/25, 10:01 PM

they all rebuild.

j0rgeOP•4/18/25, 10:01 PM

but say kinoite is red or something.

j0rgeOP•4/18/25, 10:01 PM

so bazzite sees silverblue updates, and also automerges and builds

j0rgeOP•4/18/25, 10:01 PM

but it's behind on kinoite now

j0rgeOP•4/18/25, 10:02 PM

though I guess we'll just see how it works, I am overthinking it

Robert•4/18/25, 10:02 PM

No, you're right. That could happen...

Robert•4/18/25, 10:03 PM

There's a few things we could do.

j0rgeOP•4/18/25, 10:04 PM

I know we're getting rid of it, but basically with a huge matrix like hwe has

j0rgeOP•4/18/25, 10:04 PM

I can see renovate PRing a downstream like 3 times given the churn or something.

Jj0rge bot fires off, automerges

tulip•4/18/25, 10:04 PM

i think we should gate it behind the merge queue

tulip•4/18/25, 10:04 PM

so it wont get automerged if it fails to build

Robert•4/18/25, 10:05 PM

We can stop the merge queue from pushing to the "production" tags (latest, stable etc), and re-tag the merge-queue builds as those prod tags in the push trigger.
So:

Merge queue builds all images, pushes to merge-queue only tags
Push trigger re-tags merge queue tags with production tags

Also, we can configure Renovate to group all images together, so it makes one PR for all images in a repo. This can be combined with a timer to merge things after say an hour to catch any straggling builds.

j0rgeOP•4/18/25, 10:05 PM

it does, it PRs and if checks fail it doesn't merge

Jj0rge it does, it PRs and if checks fail it doesn't merge

tulip•4/18/25, 10:05 PM

so it should be fine then, right?

Robert•4/18/25, 10:06 PM

There's things we can do, but we just need to decide what behaviour we would want

j0rgeOP•4/18/25, 10:06 PM

I like grouping them together

j0rgeOP•4/18/25, 10:07 PM

but also we could keep the production tags on the daily/weekly timer right?

j0rgeOP•4/18/25, 10:07 PM

though I guess that would negate the advantage

j0rgeOP•4/18/25, 10:08 PM

but also :latest would legit be latest.

j0rgeOP•4/18/25, 10:08 PM

let's see how this feels with main/akmods, that will be convenient right away

j0rgeOP•4/18/25, 10:09 PM

and aurora needs to follow just kinoite, and bluefin can follow silverblue, bazzite will be more complex

Robert•4/18/25, 10:10 PM

Bazzite could have separate image-versions.yaml files for each DE, and the workflows say "if image-versions.kinoite.yaml is changed, build kinoite images"

j0rgeOP•4/18/25, 10:12 PM

ooooh!

Robert•4/18/25, 10:18 PM

This is the sort of thing I did for three years at my last job.
Building out solutions to these annoying problems in GitHub and AWS.

Now I teach people why they need to rotate their passwords if they are committed to GitHub, or write reports on why patching a critical server is important

(pls hire me <big tech company>)

j0rgeOP•4/18/25, 10:21 PM

I mean those 2 things happen all the time

j0rgeOP•4/18/25, 10:21 PM

I think even at big tech companies, heh

TRUMP2028•4/18/25, 10:39 PM

Yeah at AWS we had Charlie Bell yelling at people for not rotating certs or patching servers.

Similar Threads