ci: publish the agent image on ghcr.io (MK8S-446)

.github/workflows/agent-image.yaml

@@ -0,0 +1,75 @@
+name: agent-image
+
+on:
+  workflow_call:
+    inputs:
+      is-latest:
+        description: "Also move the latest (or latest-dev) tag"
+        required: false
+        default: false
+        type: boolean
+      is-stable:
+        description: "Move latest instead of latest-dev"
+        required: false
+        default: false
+        type: boolean
+
+# One publish at a time per ref. GitHub keeps only one pending run per group
+# and cancels it when a newer one arrives, so a shared group could drop a
+# release build. Merges to main still run in order. A pre-release can race
+# with a merge to main on latest-dev: if it finishes last, latest-dev points
+# to the pre-release until the next merge that touches agent/.
+concurrency:
+  group: agent-image-${{ github.ref }}
+  cancel-in-progress: false
+
+jobs:
+  build:
+    runs-on: ubuntu-24.04
+    steps:
+      - name: Checkout
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+        with:
+          # git describe needs the tags and the history.
+          fetch-depth: 0
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@bb05f3f5519dd87d3ba754cc423b652a5edd6d2c # v4.2.0
+      - name: Log in to ghcr.io
+        uses: docker/login-action@dbcb813823bdd20940b903addbd779551569679f # v4.6.0
+        with:
+          registry: ghcr.io
+          username: ${{ github.repository_owner }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+      - name: Compute tags
+        id: tags
+        env:
+          IMAGE: ghcr.io/${{ github.repository_owner }}/image-cache-agent
+          IS_LATEST: ${{ inputs.is-latest }}
+          IS_STABLE: ${{ inputs.is-stable }}
+        run: |
+          # A tag push names the image after that tag: two tags can sit on
+          # the same commit, and describe would pick one of them.
+          if [ "$GITHUB_REF_TYPE" = "tag" ]; then
+            version="$GITHUB_REF_NAME"
+          else
+            version="$(git describe --tags --always --match='v[0-9]*')"
+          fi
+          tags="$IMAGE:$version"
+          if [ "$IS_LATEST" = "true" ]; then
+            if [ "$IS_STABLE" = "true" ]; then
+              tags="$tags,$IMAGE:latest"
+            else
+              tags="$tags,$IMAGE:latest-dev"
+            fi
+          fi
+          echo "tags=$tags" >> "$GITHUB_OUTPUT"
+      - name: Build and push
+        uses: docker/build-push-action@c3c9e263c25d99ce0380d002d59b67737d91b0dc # v7.4.0
+        with:
+          context: agent
+          # The agent only ships for amd64 (see agent/DESIGN.md).
+          platforms: linux/amd64
+          push: true
+          tags: ${{ steps.tags.outputs.tags }}
+          cache-from: type=gha,scope=image-cache-agent
+          cache-to: type=gha,mode=max,scope=image-cache-agent

.github/workflows/post-merge.yaml

@@ -0,0 +1,20 @@
+---
+name: Post Merge
+
+on:
+  push:
+    branches: [main]
+    paths:
+      - 'agent/**'
+      - '.github/workflows/agent-image.yaml'
+      - '.github/workflows/post-merge.yaml'
+
+jobs:
+  agent-image:
+    uses: ./.github/workflows/agent-image.yaml
+    permissions:
+      contents: read
+      packages: write
+    with:
+      is-latest: true
+      is-stable: false

.github/workflows/promote.yaml

@@ -8,7 +8,18 @@ on:
       - "v*"
 
 jobs:
+  # Runs before the release is created, so a broken image blocks the release.
+  build:
+    uses: ./.github/workflows/agent-image.yaml
+    permissions:
+      contents: read
+      packages: write
+    with:
+      is-latest: true
+      is-stable: ${{ ! contains(github.ref_name, '-') }}
+
   create-release:
+    needs: build
     uses: scality/workflows/.github/workflows/promote.yaml@f05282f5bd6d7e90e8536400cf468f0a54681a48 # v2.10.0
     permissions:
       contents: write

CONTRIBUTING.md

@@ -22,8 +22,10 @@ The repository holds two toolchains, and three things built from them:
 
 Keep a change inside one component when you can. Reviewers read the repository
 that way, and the toolchains have nothing in common. Releases do not follow
-that split yet: a tag cuts one version for the whole repository, and the RPMs
-are what is attached to it.
+that split yet: a tag cuts one version for the whole repository. The RPMs are
+attached to it, and the agent image is published under the same tag. A merge
+to `main` that touches `agent/` also publishes an image; the tags are listed
+in [agent/README.md](agent/README.md#deploying).
 
 ## Development environment
 

README.md

@@ -95,20 +95,21 @@ happens when an import fails, and how the package is built.
 
 ### The agent
 
-Build the image, then deploy: `deploy` applies the CRD along with the rest. No
-image is published yet, so the first step is yours.
+The image is published at `ghcr.io/scality/image-cache-agent`. Deploy it from
+a checkout of the same tag: `deploy` applies the CRD along with the rest, and
+takes them from the checkout.
 
 ```console
-make -C agent docker-build docker-push IMG=<your-registry>/image-cache-agent:<tag>
-make -C agent deploy IMG=<your-registry>/image-cache-agent:<tag>
+git checkout <tag>
+make -C agent deploy IMG=ghcr.io/scality/image-cache-agent:<tag>
 ```
 
-Those two commands are enough from an amd64 machine, deploying into a
-namespace that enforces the `privileged` Pod Security Standard. Anything else
-takes a step or two, and [agent/README.md](agent/README.md#deploying) has them:
-building for amd64 from another architecture, and the label the manifests leave
-off the namespace they create. The flags, `--cache-path` and `--resync-period`
-included, are under [Configuration](agent/README.md#configuration).
+Those commands are enough to deploy into a namespace that enforces the
+`privileged` Pod Security Standard. Anything else takes a step or two, and
+[agent/README.md](agent/README.md#deploying) has them: which image tags exist,
+building your own image, and the label the manifests leave off the namespace
+they create. The flags, `--cache-path` and `--resync-period` included, are
+under [Configuration](agent/README.md#configuration).
 
 Then declare what each node should cache:
 
@@ -246,9 +247,9 @@ Each component is built and tested independently, and both CI workflows run on
 every pull request, since a required check that never runs leaves the pull
 request waiting forever; on pushes to `main` they are scoped by path.
 
-Releases are not independent yet: a tag cuts one version for the repository,
-and the artifacts attached are the two RPMs, for both EL versions. Publishing
-the agent image is still to come.
+Releases are not independent yet: a tag cuts one version for the repository.
+It attaches the two RPMs, for both EL versions, and publishes the agent image
+under the same tag.
 
 ## Development
 

agent/AGENTS.md

@@ -104,10 +104,12 @@ Do not scaffold them back in:
   because controller-runtime defaults to `true` and would gate it the day
   someone turns leader election on. Every agent converges its own node, so
   there is no leader to elect.
-- **No helm chart.** `make build-installer` renders the manifests into
-  `dist/install.yaml`, which is gitignored and attached to nothing: today the
-  way to deploy is this checkout's `make deploy`. Publishing an artifact is a
-  separate piece of work, not a missing chart.
+- **No helm chart.** The image is published on ghcr.io (see
+  [README.md](README.md#deploying)), but the manifests are not.
+  `make build-installer` renders them into `dist/install.yaml`, which is
+  gitignored and attached to nothing: the way to deploy is this checkout's
+  `make deploy`. Publishing the manifests is a separate piece of work, not a
+  missing chart.
 
 ## Traps worth knowing
 

agent/README.md

@@ -28,8 +28,26 @@ All the commands below run from this directory.
 
 ## Deploying
 
-Build and push the image, then deploy. `deploy` builds `config/default`, which
-includes the CRD, so it covers what `install` does on its own.
+The image is published at `ghcr.io/scality/image-cache-agent`, for amd64 only:
+
+| Tag | Image |
+| --- | ----- |
+| `vX.Y.Z`, `vX.Y.Z-<pre>` | built from that release tag |
+| `latest` | the last release without a pre-release suffix to be built: re-running an older Promote moves it back |
+| `latest-dev` | the last pre-release, or the last merge to `main` that touches `agent/` |
+| `<last tag>-<n>-g<sha>` | a merge to `main` that touches `agent/`, named by `git describe --tags`: `<n>` commits after the last release tag |
+
+Deploy it from a checkout of the same tag. `deploy` builds `config/default`
+from this checkout, and only the image comes from `<tag>`: on another commit,
+the CRD and RBAC may not match the binary. `config/default` includes the CRD,
+so `deploy` covers what `install` does on its own.
+
+```console
+git checkout <tag>
+make deploy IMG=ghcr.io/scality/image-cache-agent:<tag>
+```
+
+To deploy your own build, build and push it first:
 
 ```console
 make docker-build docker-push IMG=<your-registry>/image-cache-agent:<tag>