Repository navigation
Description
Tracking issue for the continue-on-error added in #70, so it does not become permanent by default.
What was added
.github/workflows/reusable-build-iso-anaconda.yml, build job:
# arm64 source images (ghcr.io/ublue-os/bluefin:lts / lts-hwe) are currently
# missing from the published manifest lists, so arm64 legs cannot pull a
# rootfs. Soft-fail them until the upstream manifests carry arm64 again,
# so amd64 ISOs keep publishing.
continue-on-error: ${{ matrix.platform == 'arm64' }}This was the right call at the time. Every scheduled ISO run on main had been red for over a month on the same SHA (d34ce2b) — Build LTS ISOs #18, Build Stable ISOs #37, Build LTS-HWE ISOs #31, Build All ISOs #24 and Build LTS-HWE Testing ISOs #20–#25, the most recent on 2026-09-14 — and two independent causes were tangled together. Soft-failing arm64 separates them so the amd64 fix can be seen working.
Why it needs an expiry
The soft-fail has no expiry and nothing watching it. Once the upstream manifests carry arm64 again, arm64 ISO builds will keep passing silently whether or not they actually work, because the leg can no longer turn a run red. The signal that made this problem visible in the first place — red scheduled builds — is precisely what is being switched off.
So the cost of forgetting is not "arm64 stays broken". It is "arm64 breaks in some new way and nobody finds out".
Done when
Either of these, whichever comes first:
- Upstream manifests carry arm64 again —
ghcr.io/ublue-os/bluefin:ltsand:lts-hweresolve anarm64entry. Remove thecontinue-on-errorline and let the leg be load-bearing again. Blocked on bluefin-lts:stable tag resolving to arm64-only manifest — amd64 upgrades failing thallium-suite/uranium#421 / #500. - Or make the soft-fail self-removing — instead of unconditionally tolerating arm64 failure, assert that the manifest genuinely lacks an arm64 entry and fail if it ever does have one. Then the guard removes itself the moment the upstream fix lands, with no one needing to remember.
Option 2 is the better shape if arm64 is going to be some time coming. A check like:
skopeo inspect --raw docker://ghcr.io/ublue-os/bluefin:lts \
| jq -e '.manifests | map(.platform.architecture) | index("arm64") | not'
fails the moment arm64 reappears, which is the signal to delete the whole block.
Verifying it is still needed
$ skopeo inspect --raw docker://ghcr.io/ublue-os/bluefin:lts | jq '.manifests[].platform'
If arm64 is in that list, this issue is ready to close and the continue-on-error should come out.
Filed while reviewing #70.
Activity