Repository navigation
chore(deps): update ghcr.io/projectbluefin/utah-packages docker digest to cebcec0 - #311
renovate[bot] wants to merge 1 commit into
Conversation
d0820b2 to
09d8cea
Compare
73cf9cc to
509107c
Compare
509107c to
30d0081
Compare
30d0081 to
1b98e9e
Compare
1b98e9e to
2835ab5
Compare
2835ab5 to
bb71c08
Compare
99452bd to
0482c28
Compare
There was a problem hiding this comment.
The digest bump itself looks correct to me, and it is self-contained. One companion pin goes stale with it.
What I checked. server-images/Containerfile:2 is the only functional consumer of PACKAGE_IMAGE: line 3 (FROM ${PACKAGE_IMAGE} AS packages) and the bind mount at line 17 (--mount=type=bind,from=packages,source=/repository,target=/etc/utah-packages,ro) take the stage by digest, and server-images/compose.sh:6 only enables the repo by name (--enable-repo=utah-packages), so nothing else in the build depends on the literal. Checks are green (lint, tests, osquery-smoke on both arches).
Finding — stale factory pin (low, not a blocker).
server-images/packages/utah-packages.repo:1 records the digest this PR replaces:
# factory-pin: sha256:d257e97a0057e37da47995bb142c180e2352960ab13bd44594215616a395b717
Nothing keeps that line in step with the Containerfile: factory-pin occurs exactly once in the tree (no script or test asserts the two agree), and renovate.json defines no customManagers, so Renovate cannot see it either. After this merges, the .repo file documents a factory snapshot that is not the one that composed the image, and every future digest bump widens the gap. It is inert at compose time — gpgcheck=0, baseurl=file:///etc/utah-packages — so this is a provenance/auditing problem, not a build break.
Worth fixing once, outside this branch (Renovate will regenerate this one and cannot be asked to carry the fix): either add a customManagers regex entry covering server-images/packages/utah-packages.repo so the comment moves with the digest, or delete line 1 — line 2 already states the Containerfile's OCI digest is the authority, which makes the duplicate pin pure drift surface.
No action needed on this PR from me; I have not approved, labelled or merged anything.
— hive: agent=reviewer backend=omp model=anthropic/claude-opus-5:high omp=18.12.1
0482c28 to
2ae8186
Compare
2ae8186 to
fa1aef0
Compare
There was a problem hiding this comment.
Digest-only bump, and it is effective: server-images/Containerfile:2 is the sole definition of PACKAGE_IMAGE, and no workflow or justfile recipe passes a --build-arg PACKAGE_IMAGE= override, so the new digest is what FROM ${PACKAGE_IMAGE} AS packages (line 3) resolves and what the bind mount at line 17 serves to compose.sh. All 11 checks are green. Looks correct to me.
One non-blocking observation, since the build trusts this digest rather than package signatures:
Stale provenance pin — server-images/packages/utah-packages.repo:1 (severity: low)
That file opens with
# factory-pin: sha256:d257e97a0057e37da47995bb142c180e2352960ab13bd44594215616a395b717
i.e. the pre-bump digest, and this PR does not touch it. Mechanism: nothing consumes that line — it is the only factory-pin occurrence in the tree, and the deps-guard job (.github/workflows/ci.yml:114-130) only compares the dependency key set of roles/pi/files/package.json, so no check notices the divergence. Consequence: the repo definition sets gpgcheck=0 and repo_gpgcheck=0 and line 2 states the content is "Authenticated by the Containerfile's OCI digest" — which makes this comment the human-readable record of which unsigned package factory the image was composed from. After merge it names an image the build no longer uses, and it will drift again on every future digest bump.
Cheapest durable fix is a Renovate customManagers regex over that line so the two move together; renovate.json currently has no custom manager. Alternatively, delete the duplicated digest and have the comment point at ../Containerfile as the single source of truth. Either is follow-up work, not a reason to hold this PR.
Merge state is BLOCKED with every check SUCCESS, so the remaining gate is a human approval — I do not approve PRs; flagging it so a maintainer knows that is all that is left here.
— hive: agent=reviewer backend=omp model=anthropic/claude-opus-5:high omp=18.12.1
fa1aef0 to
7f994ce
Compare
6740607 to
30fcd1d
Compare
There was a problem hiding this comment.
Decision needed: @hanthor which ghcr.io/projectbluefin/utah reference should server-images/Containerfile:1 pin, now that the :latest tag returns "manifest unknown" and breaks every server-images build including master?
The diff is correct; the red check is not caused by it. This is a one-line digest bump of PACKAGE_IMAGE in server-images/Containerfile:2 (d257e97… → d51bb86…), exactly matching the body's update table, and nothing else in the tree references that digest (server-images/compose.sh:6 and server-images/packages/utah-packages.repo:3 use the repo name, not the digest, so they are unaffected).
What installer-acceptance actually failed on (job 120992756169):
[1/3] STEP 1/1: FROM ghcr.io/projectbluefin/utah-packages@sha256:d51bb860… AS packages
Trying to pull … Writing manifest to image destination <- new digest resolves fine
[2/3] STEP 1/13: FROM ghcr.io/projectbluefin/utah:latest AS common
Error: creating build container: … reading manifest latest in ghcr.io/projectbluefin/utah: manifest unknown
So the bumped digest pulled successfully; the build then died on the unrelated BASE_IMAGE default at server-images/Containerfile:1. Master HEAD fails identically — run 40134975033, same manifest unknown line — and the Home server images workflow has failed on master on every run since 2026-10-20. This is pre-existing base-branch breakage, not a regression from this PR, and it will red-flag every PR that touches server-images/.
Mechanism worth fixing separately: BASE_IMAGE is a floating :latest tag while PACKAGE_IMAGE is digest-pinned. Renovate can therefore keep the packages layer current but has no handle on the base layer, so when upstream retired the latest tag the build broke with no PR to warn about it, and no past build is reproducible either. Severity: medium — it blocks the server image pipeline entirely.
Decision needed from @hanthor: which ghcr.io/projectbluefin/utah reference should line 1 pin to (a digest, or a surviving tag such as a dated/stable channel)? I cannot pick that for you — it decides what runs on fleet machines. Note also that server-images/ is owner-owned per .github/CODEOWNERS:6, so this PR needs your review regardless of CI.
Once the base image is fixed and this branch is rebuilt, I see no reason this bump should not land. I have not merged, approved, or closed anything.
— hive: agent=reviewer backend=omp model=anthropic/claude-opus-5:high omp=18.12.1
30fcd1d to
6127541
Compare
There was a problem hiding this comment.
The diff is correct and the red check is not caused by this PR.
The change itself. server-images/Containerfile:2 moves PACKAGE_IMAGE from sha256:d257e97… to sha256:d51bb86…. It stays digest-pinned, it is the only line touched, and it is the only reference to utah-packages in the tree. Nothing else consumes that ARG except FROM ${PACKAGE_IMAGE} AS packages on line 3. Looks correct to me.
Why installer-acceptance is failing. The new digest pulls fine — the job log shows stage 1/3 completing against d51bb86…. The build then dies on the next stage, which this PR does not touch:
[2/3] STEP 1/13: FROM ghcr.io/projectbluefin/utah:latest AS common
Trying to pull ghcr.io/projectbluefin/utah:latest...
Error: creating build container: initializing source docker://ghcr.io/projectbluefin/utah:latest:
reading manifest latest in ghcr.io/projectbluefin/utah: manifest unknown
(run 40151036951, job 121042459404, step "Build starter and both node images once")
That tag comes from server-images/Containerfile:1 — ARG BASE_IMAGE=ghcr.io/projectbluefin/utah:latest — which is identical on master and unchanged by this diff. Upstream no longer publishes a latest tag for ghcr.io/projectbluefin/utah, so just server-images fails at the common stage regardless of what PACKAGE_IMAGE is set to.
Consequence, and the thing actually worth a maintainer's time. installer-acceptance is a required check, so this PR is BLOCKED on a failure it did not introduce — and so is every other PR that runs that job, until line 1 is repointed. Note the asymmetry: PACKAGE_IMAGE is digest-pinned and therefore tracked by Renovate, while BASE_IMAGE is a floating :latest that Renovate cannot see and that just disappeared from under the build. Pinning BASE_IMAGE to a digest the same way would both fix this and keep it fixed.
I have not approved, merged, or closed anything — this is a comment only. My read is that the one-line bump here is fine to land once the base-image breakage at Containerfile:1 is resolved separately.
— hive: agent=reviewer backend=omp model=anthropic/claude-opus-5:high omp=18.12.1
|
This Renovate PR has failed |
6127541 to
1d6f948
Compare
There was a problem hiding this comment.
Decision needed: @hanthor ghcr.io/projectbluefin/utah:latest now returns 404 manifest unknown and breaks installer-acceptance on master and every server-images PR — which base image reference should server-images/Containerfile:1 pin instead (a digest of a surviving upstream tag, or a different base)?
The digest bump itself is correct and safe to merge; the red check is not caused by this PR, and nothing in this branch can turn it green.
What I verified
-
server-images/Containerfile:2— the only changed line. Both digests resolve in GHCR right now: oldsha256:d257e97…→ 200, newsha256:ff6bc51…→ 200. Still a digest pin, no tag floated, matches the body exactly (d257e97→ff6bc51). The image is consumed only as a bind-mount source atserver-images/Containerfile:18-20(--mount=type=bind,from=packages,source=/repository), so the blast radius is the repo setcompose.shreads — no new layers in the shipped image. -
installer-acceptanceFAILURE is pre-existing upstream breakage. Run 40167530803, step Build starter and both node images once:Error: creating build container: initializing source docker://ghcr.io/projectbluefin/utah:latest: reading manifest latest in ghcr.io/projectbluefin/utah: manifest unknownSame step, same error on master — run 40134975033 (head
f6f3388). I queried GHCR directly:ghcr.io/projectbluefin/utah:latestreturns 404 manifest unknown. The reference comes fromserver-images/Containerfile:1(ARG BASE_IMAGE=…/utah:latest), which this PR does not touch.
Consequence for the maintainer. Renovate has automerge enabled on this PR, and installer-acceptance is a required check, so this one sits at BLOCKED forever — as will every future server-images PR and every master push — until BASE_IMAGE points at an image that exists. The queue blockage is one broken reference, not fifty bad PRs.
Secondary, pre-existing and worth fixing in the same change: PACKAGE_IMAGE is digest-pinned but BASE_IMAGE is a floating :latest (server-images/Containerfile:1). That asymmetry is what let an upstream deletion break builds with no local change, and it means the base layer of the bootc images is unverified at build time.
I have not merged, closed or labelled anything.
— hive: agent=reviewer backend=omp model=anthropic/claude-opus-5:high omp=18.12.1
1d6f948 to
c7f1d95
Compare
c7f1d95 to
93d05e5
Compare
93d05e5 to
75103c4
Compare
75103c4 to
ca3ebd6
Compare
ca3ebd6 to
e770a10
Compare
8d31ac9 to
349a57f
Compare
ae6eb2b to
9686491
Compare
This PR contains the following updates:
d257e97→cebcec0Configuration
📅 Schedule: (UTC)
🚦 Automerge: Enabled.
♻ Rebasing: Whenever PR is behind base branch, or you tick the rebase/retry checkbox.
🔕 Ignore: Close this PR and you won't be reminded about this update again.
This PR was generated by Mend Renovate. View the repository job log.