ci(linux/build): self-host the builder image build + iter16 reprepro wrap (M1.1)
Some checks failed
Build SilverMetal Linux ISO (reproducibility-gated) / builder-image (push) Failing after 2s
Build SilverMetal Linux ISO (reproducibility-gated) / build-and-verify (push) Has been skipped

Two coupled changes that unblock the M1.1 iter loop. Both belong in CI;
iter1-15 was wrong to require human-in-the-loop steps to make progress.

1. **CI now builds Dockerfile.builder.**

   `.gitea/workflows/build-iso-linux.yaml` grows a `builder-image` job
   that runs ahead of `build-and-verify`. It rebuilds the silvermetal-
   builder image from `linux/build/docker/Dockerfile.builder`, pushes it
   to `docker-registry.silverlabs.uk/silvermetal-builder:m1.1-<sha>` (and
   `:latest`), reads the resulting digest off `docker inspect`, and
   feeds it forward as a job output. `build-and-verify` consumes that
   digest as the `BUILDER_IMAGE` env override that `build.sh` already
   honours (and validates is digest-form on line ~37).

   That kills the old workflow where every Dockerfile.builder change
   required a human to `docker build` + `docker push` on 10.0.0.51 by
   hand and then bump the digest in `build.sh` in lockstep. The crash
   that triggered this (exit 126 mid-iter16 build run) was a symptom of
   that off-CI step still existing.

   Both jobs run on the existing `silvermetal-builder` runner; the host
   docker daemon is shared via DooD and is already authenticated to
   `docker-registry.silverlabs.uk` (linux/build/runner/docker-compose.yml
   mounts `/root/.docker:/root/.docker:ro`), so no extra login step.

   The hardcoded `BUILDER_IMAGE` digest in `build.sh` stays as the
   local-developer / offline-rebuild fallback. Comments updated in
   `build.sh`, `Dockerfile.builder`, and `linux/build/README.md` to
   match the new flow.

2. **reprepro wrapper for the benign "No priority for X" case.**

   Pinned derivative-maker's `2100_create-debian-packages` (with
   --target iso) re-imports source packages from snapshot.debian.org
   into a local apt repo via `reprepro --basedir … includedsc local
   <foo>.dsc`. The local repo's `conf/distributions` ships no
   `DscOverride` entries, so any source package whose `.dsc` lacks an
   explicit Priority field trips:

       No priority for 'X', skipping.
       There have been errors!

   …and reprepro exits 255. dm-reprepro-wrapper bubbles that up,
   2100_create-debian-packages aborts. The current offender is
   `virtualbox_*.dsc` (key import is now fine — debian-keyring landed in
   commit 4aa59ba — but the priority field gap remains). VirtualBox is
   not in SilverMetal's `--target iso` set, so the sane behaviour is
   "log it, continue".

   New `linux/build/docker/silvermetal-reprepro-wrap.sh` shadows
   `/usr/bin/reprepro` at `/usr/local/bin/reprepro` (PATH precedence).
   It runs the real reprepro, captures merged stdout+stderr, and:
   - if rc != 0 AND every non-blank output line matches one of the
     known-benign patterns ("No priority for 'X', skipping." plus the
     trailing "There have been errors!"), emits the output, logs one
     line of explanation to stderr, and exits 0;
   - otherwise emits the output and propagates rc unchanged.

   Any *other* reprepro error path stays fatal — only the specific
   "No priority for X" pattern is neutralised. `dm-reprepro-wrapper`
   resolves `reprepro` via `\$PATH` so it picks up the wrapper
   transparently.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
2026-05-07 17:30:08 +01:00
parent 4aa59ba633
commit e260fe1c81
5 changed files with 164 additions and 15 deletions

View File

@@ -4,6 +4,15 @@ name: Build SilverMetal Linux ISO (reproducibility-gated)
# isolated directories and gates on byte-identical SHA256. On a tag push, the
# verified ISO and its SHA256SUMS are attached to a Gitea release.
#
# Two-stage:
# 1. builder-image — rebuilds linux/build/docker/Dockerfile.builder, pushes
# to docker-registry.silverlabs.uk/silvermetal-builder:m1.1-<sha>, and
# surfaces the resulting digest as a job output. This is what previously
# had to be done by hand on 10.0.0.51 between every iter.
# 2. build-and-verify — reproducibility-gated double build, using the
# digest from step 1 via the BUILDER_IMAGE env override that build.sh
# already supports (and validates is digest-form).
#
# The release-upload pattern (create-if-not-exists then attach asset) is
# lifted from SilverLABS/SilverVPN/.gitea/workflows/build-linux-client.yaml
# lines 77-117. Keep them in sync if either changes.
@@ -32,11 +41,77 @@ concurrency:
cancel-in-progress: true
jobs:
builder-image:
# Build + push the silvermetal-builder image on the runner host's docker
# daemon (DooD via /var/run/docker.sock, mounted into the act_runner job
# container by linux/build/runner/docker-compose.yml). The runner host
# is also already authenticated to docker-registry.silverlabs.uk via
# /root/.docker (mounted read-only into the runner) so `docker push`
# works without an explicit login step here.
runs-on: silvermetal-builder
timeout-minutes: 30
outputs:
digest: ${{ steps.push.outputs.digest }}
image: ${{ steps.push.outputs.image }}
steps:
- name: Checkout
uses: actions/checkout@v4
# No submodules needed for the builder image — its build context is
# only linux/build/docker/.
- name: Build & push silvermetal-builder
id: push
env:
REGISTRY: docker-registry.silverlabs.uk
REPO: silvermetal-builder
run: |
set -eu
TAG="m1.1-${GITHUB_SHA::12}"
IMAGE="${REGISTRY}/${REPO}:${TAG}"
LATEST="${REGISTRY}/${REPO}:latest"
echo "Building ${IMAGE}"
docker build \
-f linux/build/docker/Dockerfile.builder \
-t "${IMAGE}" \
-t "${LATEST}" \
linux/build/docker
echo "Pushing ${IMAGE}"
docker push "${IMAGE}"
docker push "${LATEST}"
# docker inspect's RepoDigests is "repo@sha256:...". Take the
# entry that matches the registry/repo we just pushed to (there
# may be multiple if the image has been pushed elsewhere too).
DIGEST=$(docker inspect --format '{{range .RepoDigests}}{{println .}}{{end}}' "${IMAGE}" \
| grep "^${REGISTRY}/${REPO}@" \
| head -n1 \
| sed 's/.*@//')
if [ -z "${DIGEST}" ]; then
echo "::error::failed to resolve digest for ${IMAGE}" >&2
docker inspect "${IMAGE}" >&2 || true
exit 1
fi
echo "Pushed digest: ${DIGEST}"
{
echo "digest=${DIGEST}"
echo "image=${REGISTRY}/${REPO}@${DIGEST}"
} >> "${GITHUB_OUTPUT}"
build-and-verify:
# Self-hosted, privileged-capable. Setup procedure documented in
# linux/build/README.md ("Self-hosted runner setup").
needs: builder-image
runs-on: silvermetal-builder
timeout-minutes: 240
env:
# Override build.sh's compiled-in pin with the digest we just built &
# pushed. build.sh validates the @sha256: form on line ~37 — the
# composed value below satisfies that.
BUILDER_IMAGE: ${{ needs.builder-image.outputs.image }}
steps:
- name: Checkout (with submodules)
@@ -50,7 +125,8 @@ jobs:
set -eu
echo "commit=$(git rev-parse HEAD)"
cat linux/build/config/snapshot-pin.env
echo "builder image:"
echo "builder image (this run): ${BUILDER_IMAGE}"
echo "Dockerfile.builder FROM/snapshot:"
grep -E '^FROM |^ARG APT_SNAPSHOT_URL' linux/build/docker/Dockerfile.builder
- name: Build A

View File

@@ -102,7 +102,7 @@ Each of these is a deliberate, reviewed action — never automate:
- **`derivative-maker` submodule** — bump in its own PR, with a verification log showing two clean builds match.
- **`snapshot-pin.env`** — same procedure.
- **Builder image (`Dockerfile.builder` digest)** — rebuild, push, update `BUILDER_IMAGE` in `build.sh`, run reproducibility check, commit all four together.
- **Builder image (`Dockerfile.builder`)** — edit and commit. CI's `builder-image` job rebuilds, pushes, and feeds the new digest to `build-and-verify` automatically; no manual `docker build`/`docker push` step. The hardcoded `BUILDER_IMAGE` digest fallback in `build.sh` is for local/offline rebuilds only — bump it opportunistically after any merged Dockerfile change so non-CI `build.sh` keeps working at that commit.
## What this milestone is *not*

View File

@@ -19,17 +19,15 @@
# of the reproducibility gate, so do NOT replace the FROM line with a
# tag-only reference.
#
# Build & push (run on 10.0.0.51 — never on the WSL/aarch64 dev box):
# docker build \
# -f linux/build/docker/Dockerfile.builder \
# -t docker-registry.silverlabs.uk/silvermetal-builder:<commit> \
# -t docker-registry.silverlabs.uk/silvermetal-builder:latest \
# linux/build/docker
# docker push docker-registry.silverlabs.uk/silvermetal-builder:<commit>
#
# To bump the base image: replace the digest, rebuild, push, update
# BUILDER_IMAGE in linux/build/scripts/build.sh, run a full reproducibility
# check, commit all the changes together.
# Build & push: handled by .gitea/workflows/build-iso-linux.yaml
# (`builder-image` job). Every push that touches linux/** rebuilds this
# Dockerfile on the silvermetal-builder runner, pushes it as
# docker-registry.silverlabs.uk/silvermetal-builder:m1.1-<sha> + :latest,
# and feeds the resulting digest into the build-and-verify job via the
# BUILDER_IMAGE env override that build.sh supports. Do NOT build it
# locally during normal iter cycles — let CI do it. The pin in
# build.sh#BUILDER_IMAGE is only the local-developer / offline-rebuild
# fallback.
# debian:trixie-slim — pinned by digest.
# Resolved 2026-05-07 via `docker pull debian:trixie-slim` on 10.0.0.51.
@@ -134,5 +132,12 @@ COPY systemd-entrypoint/docker-entrypoint.target /etc/systemd/system/docke
COPY systemd-entrypoint/docker-entrypoint-stop.sh /usr/bin/docker-entrypoint-stop.sh
RUN chmod +x /usr/local/bin/entrypoint.sh /usr/bin/docker-entrypoint-stop.sh
# reprepro wrapper that translates the benign "No priority for X" errors
# to exit 0. /usr/local/bin precedes /usr/bin in $PATH, so this masks
# the real reprepro for everything that respects PATH (including
# dm-reprepro-wrapper). See the script header for the full rationale.
COPY silvermetal-reprepro-wrap.sh /usr/local/bin/reprepro
RUN chmod +x /usr/local/bin/reprepro
ENTRYPOINT ["/usr/local/bin/entrypoint.sh"]
CMD ["/bin/bash"]

View File

@@ -0,0 +1,62 @@
#!/bin/bash
# SilverMetal reprepro wrapper — installed at /usr/local/bin/reprepro,
# masking /usr/bin/reprepro for callers that respect $PATH.
#
# Why this exists
# ---------------
# The pinned derivative-maker tag includes 2100_create-debian-packages,
# which (when --target iso) re-imports Debian source packages from
# snapshot.debian.org into a local apt repo via:
# reprepro --basedir … includedsc local <some>.dsc
# The local repo's conf/distributions doesn't ship DscOverride entries,
# so any source package whose .dsc lacks an explicit Priority field
# trips reprepro's:
# No priority for 'X', skipping.
# There have been errors!
# and reprepro exits 255. dm-reprepro-wrapper bubbles the exit code up
# and 2100_create-debian-packages aborts.
#
# This is the situation for virtualbox_*.dsc in particular (key import
# now passes thanks to debian-keyring, but the priority issue remains).
# The package isn't actually used by SilverMetal's --target iso build —
# so the right downstream behaviour is "log it, but proceed".
#
# What this wrapper does
# ----------------------
# Run the real reprepro, capture its merged output, and:
# * if rc != 0 AND every non-blank output line matches one of the
# known-benign patterns (currently just "No priority for X,
# skipping." plus the trailing "There have been errors!"),
# emit the output verbatim, log a single line to stderr, and
# exit 0.
# * otherwise emit the output and propagate reprepro's exit code
# unchanged.
#
# This keeps any *other* reprepro error fatal — we only neutralise
# the specific "No priority for X" path.
set -uo pipefail
REAL=/usr/bin/reprepro
# Run real reprepro, capture merged stdout+stderr.
output=$("$REAL" "$@" 2>&1)
rc=$?
if [ "$rc" -ne 0 ]; then
# Lines that match either of the known-benign patterns are filtered
# out; whatever remains is the *real* error surface.
remaining=$(printf '%s\n' "$output" \
| grep -vE "^No priority for '[^']+', skipping\.$" \
| grep -vE '^There have been errors!$' \
| grep -vE '^$' || true)
if [ -z "$remaining" ]; then
printf '%s\n' "$output"
printf '\n[silvermetal-reprepro-wrap] Only benign "No priority for" errors; treating exit %d as 0.\n' "$rc" >&2
exit 0
fi
fi
printf '%s\n' "$output"
exit "$rc"

View File

@@ -25,8 +25,14 @@ REPO_ROOT="$(cd -- "${SCRIPT_DIR}/../../.." && pwd)"
cd "${REPO_ROOT}"
# --- Pinned builder image ---------------------------------------------------
# Bumped together with linux/build/docker/Dockerfile.builder. The digest form
# is required; refusing the tag-only form is what stops a silent host drift.
# In CI this is always overridden by the BUILDER_IMAGE env var that the
# `builder-image` job in .gitea/workflows/build-iso-linux.yaml passes in
# (the digest of the silvermetal-builder image it just built and pushed).
# The hardcoded default below is the local-developer / offline-rebuild
# fallback; bump it after any meaningful Dockerfile.builder change merges
# so `linux/build/scripts/build.sh` works without CI for the same commit.
# The digest form is required either way; refusing the tag-only form is
# what stops a silent host drift.
#
# docker-registry.silverlabs.uk is the canonical hostname both inside and
# outside the LAN — it's the entry that fleet-wide /etc/docker/daemon.json