@antheas another error with rechunk popped up this time with ucore https://github.com/m2Giles/m2os/a

@antheas another error with rechunk popped up this time with ucore
https://github.com/m2Giles/m2os/actions/runs/10335823416/job/28610865502#step:18:615

M2OP•8/11/24, 1:28 AM

There is no previous image for that image.

M2OP•8/11/24, 2:10 AM

Think I figured out the cause. You put GCC in the same layer and it's not on image yet

M2OP•8/11/24, 2:34 AM

Nope not it.

M2OP•8/11/24, 2:50 AM

looks like its gcc just failing to install correctly

M2OP•8/11/24, 3:42 AM

okay. I'm guessing its somehow having a mixed subpackages of gcc causing the issue

Jj0rge yep, watching it

Ivo•8/11/24, 4:38 AM

We were able to fix our issue. /etc/ instead of /usr/etc/ in a script and it builds. I am uncertain if this change was introduced with rechunks. Thx

antheas•8/11/24, 5:58 AM

Latest ver woes I think

antheas•8/11/24, 5:59 AM

Actually dunno

antheas•8/11/24, 5:59 AM

Can't see the error from my phone I'll look at it in a bit

MM2 @antheas another error with rechunk popped up this time with ucore https://githu...

antheas•8/11/24, 7:03 AM

ill look into it

antheas•8/11/24, 7:11 AM

pulled the image and i get the error

MM2 okay. I'm guessing its somehow having a mixed subpackages of gcc causing the iss...

antheas•8/11/24, 7:39 AM

https://github.com/hhd-dev/rechunk/releases/tag/v0.8.4

GitHub

Release v0.8.4: Packaging fixup · hhd-dev/rechunk

Fixes a bug in the prefill algorithm that caused some packages to not be allocated and a subsequent crash. In addition, makes it so if this happens again due to a different bug an error is emitted ...

antheas•8/11/24, 7:45 AM

id suggest a smaller number of layers for ucore

antheas•8/11/24, 7:46 AM

try 40?

antheas•8/11/24, 7:46 AM

or maybe 30

antheas•8/11/24, 7:46 AM

30 seems good

antheas•8/11/24, 7:47 AM

i used 69 so they'd be 70 but ostree-rs-ext adds 2 more and they became 71

antheas•8/11/24, 7:49 AM

20 is also not bad

antheas•8/11/24, 7:51 AM

          - 61.466 MB /usr/libexec/docker/cli-plugins/docker-compose                                                                             
          - 53.330 MB /usr/libexec/docker/cli-plugins/docker-buildx

          - 61.466 MB /usr/libexec/docker/cli-plugins/docker-compose                                                                             
          - 53.330 MB /usr/libexec/docker/cli-plugins/docker-buildx

antheas•8/11/24, 7:54 AM

https://github.com/hhd-dev/rechunk/releases/tag/v0.8.5

GitHub

Release v0.8.5: Docker cli packages · hhd-dev/rechunk

Adds meta packages for docker compose and docker buildx, which are large executables of around 60mb.
Full Changelog: v0.8.4...v0.8.5

Ion•8/11/24, 12:09 PM

Hi @Kyle Gospo, I don't understand this commit: https://github.com/ublue-os/bluefin/pull/1580. What is "undefined behavior"? Sorry to bother you, but that PR broke my custom build and I want to understand why it happend

Robert•8/11/24, 12:12 PM

Having /usr/etc merge into /etc wasn't documented, and wasn't designed to be used as we did. From what I understand, it was an internal functionality of rpm-ostree, meant to be used for non-OCI workflows and/or when layering locally.

It also started causing issues when we were rechunking our OCI images using the hhd-dev/rechunk action

xyny•8/11/24, 12:13 PM

writing to /usr/etc/ at build time is apparently undocumented and undefined behavior with rpm-ostree, and writing to /etc/ does the same thing, we at bluebuild wrote an announcement of this, but it's still unreleased, there's a couple things to do first: https://github.com/blue-build/website/pull/65/files

RRobert Having /usr/etc merge into /etc wasn't documented, and wasn't designed to be use...

antheas•8/11/24, 12:29 PM

yes and no. Essentially the code was written assuming only /etc exists, which caused them to merge during deployment. This merge was a byproduct. If you look at the podman PR for bootc, it will try to move /etc to /usr/etc and fail because it exists. Therefore we slowly need to start migrating

antheas•8/11/24, 12:30 PM

however the keyword here is SLOWLY. I want to have full backward compatibility with original ublue images prior to this. This includes merging /usr/etc to /etc the exact same way rpm-ostree would

antheas•8/11/24, 12:30 PM

and if there are any deployment issues with rechunk and /usr/etc existing I promise to fix them. I already fixed most of them.

antheas•8/11/24, 12:32 PM

if ublue starts breaking downstream images because of this, imo you should consider reverting the PR and letting rechunk handle it until we can define a behavior for /usr/etc existing

Aantheas and if there are any deployment issues with rechunk and /usr/etc existing I prom...

Robert•8/11/24, 12:33 PM

I didn't mean to point blame towards the rechunk action. I just meant rechunk didn't expect this because it's not the way rpm-ostree expects, and therefore it didn't work immediately. This sparked the conversation about how we should be handling config files in the future.

antheas•8/11/24, 12:34 PM

Of course no offense. It is a bit of undefined behavior that you came to expect with your images. So lets keep it that way and move slowly instead. In my opinion

antheas•8/11/24, 12:34 PM

there is no rush, we can tweak the merging in the action as necessary until everything works as expected

antheas•8/11/24, 12:35 PM

In the meantime, lets setup a communication channel with the redhat folk so that bootc can also be prepared to handle the quirks we found

antheas•8/11/24, 12:35 PM

https://github.com/hhd-dev/rechunk/issues/2 i started this issue to track this

GitHub

OCI Image boot quirks or "The road to `FROM fedora: 40`" · Issue #2...

Lets use this issue to track quirks inherent to booting an OCI image. Hopefully, when this issue closes it will be possible for someone to boot an image made with FROM fedora:40 First lets begin wi...

antheas•8/11/24, 12:37 PM

Also the last version of rechunk will also log all the files it finds under /usr/etc to ease this migration

antheas•8/11/24, 12:37 PM

but again theres no rush here

antheas•8/11/24, 12:37 PM

we are not anywhere close to migrating to bootc under podman. Until then, both bootc and rpm-ostree will have the same behavior so you can still rely on it

antheas•8/11/24, 12:39 PM

I was not notified to it by Colins. I came to the realization by reading the ostree source code across all libraries when building rechunk

antheas•8/11/24, 12:39 PM

And in that effort, I had ostree panic and not boot because both dirs existed, then had to merge them, then saw how rpm-ostree and ostree-rs-ext handled it etc.

Aantheas I was not notified to it by Colins. I came to the realization by reading the ost...

fiftydinar•8/11/24, 12:42 PM

I will edit the blog to mention that Collins notified Ublue folks only

antheas•8/11/24, 12:43 PM

oh so its official

antheas•8/11/24, 12:43 PM

well in that case

antheas•8/11/24, 12:43 PM

lol

antheas•8/11/24, 12:43 PM

well good thing he did

antheas•8/11/24, 12:43 PM

nobody notified me lmao

Robert•8/11/24, 12:43 PM

I've not heard anything from Colin directly. Did I miss something?

fiftydinar•8/11/24, 12:43 PM

I mean, Kyle mentioned that Colins told him that

antheas•8/11/24, 12:44 PM

:clueless:

fiftydinar•8/11/24, 12:45 PM

Or he maybe got informed indirectly from Colins, idk

antheas•8/11/24, 12:46 PM

lets ask Kyle

IIon Hi @Kyle Gospo, I don't understand this commit: https://github.com/ublue-os/blue...

j0rge•8/11/24, 1:26 PM

what broke in your build?

Jj0rge what broke in your build?

Ion•8/11/24, 1:35 PM

#!/usr/bin/env bash

# Tell build process to exit if there are any errors.
set -oue pipefail

rm /usr/etc/skel/.config/autostart/bluefin-firstboot.desktop
rm /usr/etc/yafti.yml
rm /usr/etc/profile.d/bluefin-firstboot.sh

#!/usr/bin/env bash

# Tell build process to exit if there are any errors.
set -oue pipefail

rm /usr/etc/skel/.config/autostart/bluefin-firstboot.desktop
rm /usr/etc/yafti.yml
rm /usr/etc/profile.d/bluefin-firstboot.sh

That was one of the scripts that exited with an error so the build action was failing. Apologies for not being more specific. My images wouldn't get built, not that they "broke".

- 61.466 MB /usr/libexec/docker/cli-plugins/docker-compose - 53.330 MB /usr/libexec/docker/cli-plugins/docker-buildx

#!/usr/bin/env bash # Tell build process to exit if there are any errors. set -oue pipefail rm /usr/etc/skel/.config/autostart/bluefin-firstboot.desktop rm /usr/etc/yafti.yml rm /usr/etc/profile.d/bluefin-firstboot.sh

@antheas another error with rechunk popped up this time with ucore https://github.com/m2Giles/m2os/a

Similar Threads

Similar Threads

Similar Threads