The Android job ran in gitea.tourolle.paris/dtourolle/darkroom-android:latest, a tag that had never been pushed. The image existed only as a local darkroom-android:latest on one machine, so every Android job died at docker pull with "manifest unknown" before reaching a step. The registry API confirms it: that manifest is a 404 while the other builder images answer 200. android-image.yml now builds and pushes it, following KPN's docker.yaml — host runner rather than a container, so it has the Docker daemon and the host's cached registry credentials, and a plain-git checkout because that host has no Node for actions/checkout. Where it diverges from KPN: that workflow gates on dorny/paths-filter running inside the builder image, which here would need the very image that is missing. The tag is the git tree hash of docker/android instead, which changes when and only when a file there changes. An unrelated push reuses the image, a Dockerfile edit cannot keep serving a stale latest, and a missing tag rebuilds itself without a manual step. The presence probe is curl against the registry API, not `docker manifest inspect`. The latter exits 1 on this registry even for tags that are plainly there — jellytau-builder:latest answers HTTP 200 while docker reports "manifest unknown" for it — and trusting it would have rebuilt 7 GB on every push. The HEAD request also yields Docker-Content-Digest, so latest is repointed only when the digests actually disagree, without pulling any layers. A probe that cannot authenticate falls through to building. Rebuilding when it was unnecessary costs minutes; skipping a build that was needed is the failure this commit exists to remove. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
171 lines
7.7 KiB
YAML
171 lines
7.7 KiB
YAML
name: '🐳 Android image'
|
|
|
|
# Builds and pushes gitea.tourolle.paris/dtourolle/darkroom-android, the job
|
|
# container for the Android leg of build-and-test.yml.
|
|
#
|
|
# It exists because that image previously lived only on a developer's laptop:
|
|
# the workflow referenced a tag that had never been pushed, and every Android
|
|
# job died at `docker pull` with "manifest unknown" before running a step. The
|
|
# image is now reproducible from the repo rather than from one machine.
|
|
#
|
|
# Called by build-and-test.yml on every push, and runnable by hand via
|
|
# workflow_dispatch. It is cheap when nothing changed — see the guard below.
|
|
on:
|
|
workflow_call:
|
|
inputs:
|
|
force:
|
|
description: 'Rebuild even if the registry already has this image ("true"/"false")'
|
|
type: string
|
|
default: 'false'
|
|
workflow_dispatch:
|
|
inputs:
|
|
force:
|
|
description: 'Rebuild even if the registry already has this image ("true"/"false")'
|
|
type: string
|
|
default: 'false'
|
|
|
|
# Gitea's act_runner mangles boolean workflow inputs passed through an
|
|
# expression — they arrive as false regardless of what was sent. Every input
|
|
# here is a string compared with == 'true', as in KPN's docker.yaml.
|
|
|
|
env:
|
|
IMAGE: gitea.tourolle.paris/dtourolle/darkroom-android
|
|
|
|
jobs:
|
|
build:
|
|
runs-on: linux/amd64
|
|
name: Build and push
|
|
# Deliberately NOT in a container: this job needs the host Docker daemon to
|
|
# build an image, and the host's cached ~/.docker/config.json to push it.
|
|
# That is also why there is no `docker login` step — the runner host was
|
|
# authenticated to the registry during setup.
|
|
|
|
steps:
|
|
# The host has no Node, so the JS-based actions/checkout cannot run here.
|
|
# A minimal shallow fetch with plain git gets the same tree.
|
|
- name: Checkout
|
|
run: |
|
|
set -e
|
|
git init -q .
|
|
git remote add origin "${{ github.server_url }}/${{ github.repository }}.git"
|
|
git -c http.extraheader="AUTHORIZATION: basic $(printf '%s' '${{ github.actor }}:${{ github.token }}' | base64 -w0)" \
|
|
fetch --depth 1 origin "${{ github.sha }}"
|
|
git checkout -q FETCH_HEAD
|
|
|
|
# The image is tagged by the content of docker/android, not by the commit
|
|
# that happened to touch it. `git rev-parse HEAD:<dir>` is the tree object
|
|
# id — it changes when and only when a file in that directory changes, so
|
|
# an unrelated push reuses the existing image and a Dockerfile edit can
|
|
# never silently keep serving a stale `latest`.
|
|
#
|
|
# Using the commit sha instead would rebuild 7 GB on every push; using a
|
|
# paths-filter action would need a container that has Node, and the only
|
|
# one this repo would reach for is the very image being built.
|
|
- name: Resolve image tag
|
|
id: tag
|
|
run: |
|
|
set -e
|
|
TREE=$(git rev-parse HEAD:docker/android)
|
|
echo "tree=$TREE" >> "$GITHUB_OUTPUT"
|
|
echo "docker/android tree: $TREE"
|
|
|
|
# Skip the build when the registry already holds this exact content. This
|
|
# is what keeps the job a few seconds long on a normal push, and what
|
|
# makes it self-healing: if the tag is missing for any reason, including
|
|
# the image having never been pushed at all, it gets built here.
|
|
#
|
|
# The probe is curl against the registry API, NOT `docker manifest
|
|
# inspect`. The latter exits 1 on this registry even for tags that are
|
|
# demonstrably present — jellytau-builder:latest answers HTTP 200 to the
|
|
# API while `docker manifest inspect` reports "manifest unknown" for it.
|
|
# Trusting that would have rebuilt 7 GB on every single push.
|
|
#
|
|
# A HEAD request also gives the digest for free, which is how the repoint
|
|
# decision below is made without pulling any layers.
|
|
- name: Query registry
|
|
id: check
|
|
env:
|
|
# The runner's own credentials, so this does not depend on how the
|
|
# host's ~/.docker/config.json happens to be set up.
|
|
REG_USER: ${{ github.actor }}
|
|
REG_PASS: ${{ github.token }}
|
|
TREE: ${{ steps.tag.outputs.tree }}
|
|
run: |
|
|
set -eu
|
|
ACCEPT='application/vnd.oci.image.index.v1+json,application/vnd.docker.distribution.manifest.v2+json,application/vnd.oci.image.manifest.v1+json,application/vnd.docker.distribution.manifest.list.v2+json'
|
|
API="https://gitea.tourolle.paris/v2/dtourolle/darkroom-android/manifests"
|
|
|
|
# Prints "<http-status> <digest-or-empty>" for a tag.
|
|
probe() {
|
|
curl -sI -u "$REG_USER:$REG_PASS" -H "Accept: $ACCEPT" "$API/$1" \
|
|
| tr -d '\r' \
|
|
| awk 'BEGIN{s="000";d=""} /^HTTP/{s=$2} tolower($1)=="docker-content-digest:"{d=$2} END{print s, d}'
|
|
}
|
|
|
|
read -r TREE_STATUS TREE_DIGEST <<EOF
|
|
$(probe "$TREE")
|
|
EOF
|
|
read -r LATEST_STATUS LATEST_DIGEST <<EOF
|
|
$(probe latest)
|
|
EOF
|
|
|
|
echo "tag $TREE -> HTTP $TREE_STATUS ${TREE_DIGEST:-(no digest)}"
|
|
echo "tag latest -> HTTP $LATEST_STATUS ${LATEST_DIGEST:-(no digest)}"
|
|
|
|
# Build unless the registry definitively confirms this content is
|
|
# already there. An auth failure or an unreachable registry lands
|
|
# here too, and rebuilding needlessly is the safe direction to fail —
|
|
# skipping a build that was needed is what breaks the Android job.
|
|
if [ "${{ inputs.force }}" = "true" ]; then
|
|
echo "forced rebuild requested"
|
|
echo "build=true" >> "$GITHUB_OUTPUT"
|
|
echo "repoint=false" >> "$GITHUB_OUTPUT"
|
|
elif [ "$TREE_STATUS" != "200" ]; then
|
|
echo "registry does not have this content — building"
|
|
echo "build=true" >> "$GITHUB_OUTPUT"
|
|
echo "repoint=false" >> "$GITHUB_OUTPUT"
|
|
elif [ -n "$TREE_DIGEST" ] && [ "$TREE_DIGEST" = "$LATEST_DIGEST" ]; then
|
|
echo "registry is already correct — nothing to do"
|
|
echo "build=false" >> "$GITHUB_OUTPUT"
|
|
echo "repoint=false" >> "$GITHUB_OUTPUT"
|
|
else
|
|
echo "content is present but latest points elsewhere — repointing"
|
|
echo "build=false" >> "$GITHUB_OUTPUT"
|
|
echo "repoint=true" >> "$GITHUB_OUTPUT"
|
|
fi
|
|
|
|
# Context is docker/android, matching the README's build command. The
|
|
# Dockerfile COPYs nothing from the repo, so it needs no wider context —
|
|
# and a narrow context keeps the daemon from tarring up the whole tree,
|
|
# target/ included.
|
|
- name: Build
|
|
if: ${{ steps.check.outputs.build == 'true' }}
|
|
run: |
|
|
set -e
|
|
docker build \
|
|
-t "$IMAGE:${{ steps.tag.outputs.tree }}" \
|
|
-t "$IMAGE:latest" \
|
|
docker/android
|
|
|
|
# Both tags are pushed: the tree tag is what the guard above looks for on
|
|
# the next run, and `latest` is what build-and-test.yml pulls.
|
|
- name: Push
|
|
if: ${{ steps.check.outputs.build == 'true' }}
|
|
run: |
|
|
set -e
|
|
docker push "$IMAGE:${{ steps.tag.outputs.tree }}"
|
|
docker push "$IMAGE:latest"
|
|
|
|
# A cache hit on the tree tag says nothing about where `latest` points — a
|
|
# reverted Dockerfile or a build from another branch can leave it on
|
|
# different content. This runs only when the digests above actually
|
|
# disagree, so the common case costs nothing; the layers are already in
|
|
# the registry, so the push that follows uploads a manifest, not 7 GB.
|
|
- name: Repoint latest
|
|
if: ${{ steps.check.outputs.repoint == 'true' }}
|
|
run: |
|
|
set -e
|
|
docker pull "$IMAGE:${{ steps.tag.outputs.tree }}"
|
|
docker tag "$IMAGE:${{ steps.tag.outputs.tree }}" "$IMAGE:latest"
|
|
docker push "$IMAGE:latest"
|