From 65067e241f90206f2d32b3fcbf42ab402f31a48f Mon Sep 17 00:00:00 2001 From: kylerankin Date: Sun, 20 Sep 2026 08:15:24 +0000 Subject: [PATCH 1/4] fix(configure-services): stop claiming Utah images are rechunked scripts/configure-services.sh says Utah rechunks with chunkah through projectbluefin/actions' reusable-build. It does not: the reusable workflow's testing-stream guard skips the rechunk (and SBOM) steps, and Utah publishes only the testing stream, so its images are never rechunked (#131). The rechunker-group-fix.service unit is still removed, but for the reason the next paragraph already gives -- it breaks the boot via a systemd job cycle -- not because of any rechunk claim. Correct the comment so it stops asserting a behavior the build does not perform; the real fix needs a reusable-workflow input (#131) and lives in projectbluefin/actions. Assisted-by: Pi (lemonade/Ornith-1.5-35B-A3B-GGUF-Q6_K) Signed-off-by: kylerankin --- scripts/configure-services.sh | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/scripts/configure-services.sh b/scripts/configure-services.sh index 53f684b4e..1121f3d58 100644 --- a/scripts/configure-services.sh +++ b/scripts/configure-services.sh @@ -37,9 +37,9 @@ disable_unit() { # Hummingbird intentionally does not ship every Bluefin integration package. # Not enabled: rechunker-group-fix.service. It exists to repair the # /usr/lib/{group,gshadow} damage left by the legacy rechunker -# (hhd-dev/rechunk) when rebasing to an image built without it. Utah rechunks -# with chunkah, through projectbluefin/actions' reusable-build, and has never -# been through legacy-rechunk -- so there is nothing here for it to repair. +# (hhd-dev/rechunk) when rebasing to an image built without it. Utah has never +# run the legacy rechunker, so this repair is unnecessary for it -- but that is +# not why the unit is removed. # # It is not merely useless, it breaks the boot. The unit orders itself both # After=local-fs.target and Before=systemd-sysusers.service, and local-fs From 726f8cafdf3e91730f6aa3c9480d17b24743c3c8 Mon Sep 17 00:00:00 2001 From: kylerankin Date: Sun, 20 Sep 2026 09:35:54 +0000 Subject: [PATCH 2/4] fix(docs): stop delegating rechunking to projectbluefin/actions@v1 ci-workflows.md:39 still listed rechunking as delegated to projectbluefin/actions@v1 -- the same false claim configure-services.sh #211 removes. Utah publishes only the testing stream, and the reusable workflow's testing-stream guard skips the rechunk step (#131), so its images are never rechunked. Correct the doc in the same pass. Addresses the low docs-currency finding in projectbluefin/utah#211. --- docs/skills/ci-workflows.md | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/docs/skills/ci-workflows.md b/docs/skills/ci-workflows.md index 017b713d9..75ea2f5cd 100644 --- a/docs/skills/ci-workflows.md +++ b/docs/skills/ci-workflows.md @@ -36,9 +36,14 @@ each pinned to a SHA tagged `v1`: explicitly dispatch this, or manually supply a successful testing build run ID. CI delegates builds, vulnerability reporting, SBOMs, keyless signatures, -provenance, caching, and rechunking to `projectbluefin/actions@v1` (originated +provenance, and caching to `projectbluefin/actions@v1` (originated as a `docs/building.md` design bullet; now lives in this skill). +Note: rechunking is **not** delegated. Utah publishes only the testing stream, +and the reusable workflow's testing-stream guard skips the rechunk (and SBOM) +steps (#131), so Utah images are never rechunked -- the same claim this PR +removes from `scripts/configure-services.sh`. + ## contract: the cheap gate `build.yml` opens with a container-based gate so a contract package that none of From e770c3ce661958382ea7280f12dae55837acc55e Mon Sep 17 00:00:00 2001 From: "sec-check[bot]" Date: Sun, 20 Sep 2026 06:31:30 -0400 Subject: [PATCH 3/4] fix(docs): drop SBOM from delegated list and the self-referential PR mention The testing-stream guard in projectbluefin/actions@v1 skips both the rechunk and SBOM steps (#131), and Utah publishes only the testing stream, so SBOMs are never produced. Keeping "SBOMs" in the delegated list contradicted the note directly beneath it. Also replace the "this PR" reference, which goes stale once merged, with the standing fact. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> --- docs/skills/ci-workflows.md | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-) diff --git a/docs/skills/ci-workflows.md b/docs/skills/ci-workflows.md index 75ea2f5cd..9c949ac56 100644 --- a/docs/skills/ci-workflows.md +++ b/docs/skills/ci-workflows.md @@ -35,14 +35,15 @@ each pinned to a SHA tagged `v1`: - `.github/workflows/post-testing-e2e.yml` -- successful non-PR testing builds explicitly dispatch this, or manually supply a successful testing build run ID. -CI delegates builds, vulnerability reporting, SBOMs, keyless signatures, +CI delegates builds, vulnerability reporting, keyless signatures, provenance, and caching to `projectbluefin/actions@v1` (originated as a `docs/building.md` design bullet; now lives in this skill). -Note: rechunking is **not** delegated. Utah publishes only the testing stream, -and the reusable workflow's testing-stream guard skips the rechunk (and SBOM) -steps (#131), so Utah images are never rechunked -- the same claim this PR -removes from `scripts/configure-services.sh`. +Note: rechunking and SBOM generation are **not** delegated. Utah publishes only +the testing stream, and the reusable workflow's testing-stream guard skips both +the rechunk and the SBOM steps (#131), so Utah images are never rechunked and no +SBOM is ever produced. `scripts/configure-services.sh` therefore does not claim +the image is rechunked. ## contract: the cheap gate From fc3bcdf68345dc250269a2cb741d9d5fe8d6234a Mon Sep 17 00:00:00 2001 From: Copilot <223556219+Copilot@users.noreply.github.com> Date: Sun, 20 Sep 2026 07:18:39 -0400 Subject: [PATCH 4/4] fix(packages): drop the last rechunk claim from the CUDA note The CUDA size argument in packages/utah.toml still said the closure would have to survive an OCI export and a rechunk on a runner. Utah publishes only the testing stream, where the reusable workflow skips the rechunk step (#131), so no Utah image is ever rechunked. State the real cost -- building and pushing from a runner that already freed disk for the kernel build -- so the manifest matches the rest of the cleanup. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> --- packages/utah.toml | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/packages/utah.toml b/packages/utah.toml index 3f3a5778e..fc235175f 100644 --- a/packages/utah.toml +++ b/packages/utah.toml @@ -268,8 +268,8 @@ packages = [ # available. Measured from its metadata rather than guessed: the cuda-toolkit # closure is 65 packages, 3.97 GB to download and 7.68 GB installed. This image # is a few GB, so that multiplies it, in every pull and every layer, and it has -# to survive an OCI export and a rechunk on a runner that already needed disk -# freed for the kernel build. +# to be built and pushed from a runner that already needed disk freed for the +# kernel build. # # The full `cuda` metapackage is worse and is not even installable here: it # requires kmod-nvidia-open-dkms, nvidia-driver-libs, nvidia-driver-common and