feat(release): release-please tagging, version-tagged images, Renovate

.github/workflows/build-and-publish.yml

@@ -1,5 +1,5 @@
-# Builds + publishes container images to ghcr.io with `:latest` and
-# `:<sha>` tags. Cluster rollout is handled out-of-band: Keel polls
+# Builds + publishes container images to ghcr.io with `:latest`,
+# `:<sha>`, and release tags. Cluster rollout is handled out-of-band: Keel polls
 # ghcr every 2 minutes and rolls the matching Deployment when the
 # `:latest` digest changes, while Flux reconciles manifest changes
 # from `main`. Neither is this workflow's responsibility — the name
@@ -9,6 +9,7 @@ name: Build & Publish
 on:
   push:
     branches: [main]
+    tags: ['v*.*.*']
   # Manual trigger for the cases where a push event never fires the
   # workflow (rare GitHub oddity — observed once with the `#308`
   # merge: workflow silently absent for that SHA) or where a merge
@@ -52,7 +53,7 @@ jobs:
       - uses: dorny/paths-filter@v4.0.1
         id: filter
         with:
-          base: ${{ github.event.before }}
+          base: ${{ github.ref_type == 'tag' && github.sha || github.event.before }}
           filters: |
             workflow:
               - '.github/workflows/build-and-publish.yml'
@@ -136,11 +137,8 @@ jobs:
           #
           # Multi-arch is gated on `main` only. The arm64 leg builds under
           # QEMU emulation and roughly doubles the UI build time; we don't
-          # want to pay that on every reconcile / dispatch / future trigger
-          # that someone might add. Today the workflow's `on:` only fires
-          # for push to main, but this guard makes the intent explicit and
-          # forward-safe — if you add workflow_dispatch later, feature
-          # branches still build amd64-only.
+          # want to pay that on every reconcile, dispatch, or release-tag
+          # trigger; feature branches and tags still build amd64-only.
           AMD64=linux/amd64
           if [[ "${{ github.ref }}" == "refs/heads/main" && "${{ github.event_name }}" == "push" ]]; then
             UI_PLATFORMS=linux/amd64,linux/arm64
@@ -150,8 +148,12 @@ jobs:
           fi
 
           # `force_all=true` on workflow_dispatch overrides every
-          # per-service filter so all images rebuild in one go.
+          # per-service filter so all images rebuild in one go. Release tags
+          # also rebuild every service so each image gets the version tag.
           FORCE_ALL="${{ github.event.inputs.force_all || 'false' }}"
+          if [[ "$GITHUB_REF_TYPE" == "tag" && "$GITHUB_REF_NAME" == v*.*.* ]]; then
+            FORCE_ALL=true
+          fi
 
           [[ "$(any "$FORCE_ALL" "$KOTLIN_SHARED"   "$FILTER_AUTH_API")"      == "true" ]] && append auth-api      services/auth-api/Dockerfile      "$AMD64"
           [[ "$(any "$FORCE_ALL" "$KOTLIN_SHARED"   "$FILTER_ASSISTANT_API")" == "true" ]] && append assistant-api services/assistant-api/Dockerfile "$AMD64"
@@ -197,6 +199,14 @@ jobs:
       - uses: actions/checkout@v6.0.2
       - name: Set image prefix (lowercase)
         run: echo "IMAGE_PREFIX=$(echo '${{ github.repository }}' | tr '[:upper:]' '[:lower:]')" >> "$GITHUB_ENV"
+      - name: Compute release version
+        id: release-version
+        run: |
+          if [[ "$GITHUB_REF_TYPE" == "tag" && "$GITHUB_REF_NAME" == v*.*.* ]]; then
+            echo "version=${GITHUB_REF_NAME#v}" >> "$GITHUB_OUTPUT"
+          else
+            echo "version=" >> "$GITHUB_OUTPUT"
+          fi
       # QEMU is only needed when a service builds for a non-runner arch
       # (the UIs build linux/arm64 alongside amd64). No-op for the
       # amd64-only JVM services.
@@ -225,5 +235,6 @@ jobs:
           tags: |
             ${{ env.REGISTRY }}/${{ env.IMAGE_PREFIX }}/${{ matrix.service.name }}:${{ github.sha }}
             ${{ env.REGISTRY }}/${{ env.IMAGE_PREFIX }}/${{ matrix.service.name }}:latest
+            ${{ steps.release-version.outputs.version && format('{0}/{1}/{2}:{3}', env.REGISTRY, env.IMAGE_PREFIX, matrix.service.name, steps.release-version.outputs.version) || '' }}
           cache-from: type=gha,scope=${{ matrix.service.name }}
           cache-to: type=gha,mode=max,scope=${{ matrix.service.name }}

.github/workflows/release.yml

@@ -0,0 +1,23 @@
+name: Release
+
+on:
+  push:
+    branches: [main]
+
+permissions:
+  contents: write
+  pull-requests: write
+
+concurrency:
+  group: release-please
+  cancel-in-progress: false
+
+jobs:
+  release:
+    name: Release Please
+    runs-on: ubuntu-latest
+    steps:
+      - uses: googleapis/release-please-action@v4
+        with:
+          config-file: release-please-config.json
+          manifest-file: .release-please-manifest.json

.release-please-manifest.json

@@ -0,0 +1,3 @@
+{
+  ".": "0.1.0"
+}

docs/runbooks/versioned-deploy.md

@@ -0,0 +1,36 @@
+# Versioned Deploy Model
+
+Production deploys are driven by explicit version pins in Git, not by pushes to
+`main` or mutable `:latest` tags.
+
+## Target Model
+
+- `main` is the integration branch. Merging to `main` does not by itself define
+  the production version.
+- release-please creates release tags in the form `vX.Y.Z`.
+- `build-and-publish` publishes in-house container images with version tags that
+  match the release tag.
+- A release PR bumps explicit image tags under `platform/cluster/flux/**` from
+  the previous version to the new version.
+- Flux deploys the commit that pins those image tags.
+- Rollback is a Git revert of the release PR that bumped the image tags.
+
+## Deploying a Version
+
+To deploy a specific release, reconcile the commit that pins the desired
+`vX.Y.Z` image tags in `platform/cluster/flux/**`. The deployed state is the
+Flux-applied Git state, so the image tags in the cluster manifests are the
+source of truth.
+
+## Rollback
+
+Rollback is performed by reverting the tag-bump commit or PR. Flux reconciles
+the reverted manifests and returns the affected workloads to the previously
+pinned image versions.
+
+## Keel Removal Sequencing
+
+Keel-based `:latest` auto-rolls are removed from the in-house app deploy path in
+a separate, sequenced PR. That Flux/Keel change must land only after the first
+versioned in-house images have been published, so every workload can be pinned
+to an existing version tag before `:latest` automation is removed.

release-please-config.json

@@ -0,0 +1,10 @@
+{
+  "packages": {
+    ".": {
+      "release-type": "simple",
+      "bump-minor-pre-major": true,
+      "changelog-path": "CHANGELOG.md",
+      "include-component-in-tag": false
+    }
+  }
+}

renovate.json

@@ -0,0 +1,22 @@
+{
+  "$schema": "https://docs.renovatebot.com/renovate-schema.json",
+  "extends": ["config:recommended", ":semanticCommits", ":dependencyDashboard"],
+  "rangeStrategy": "pin",
+  "labels": ["dependencies"],
+  "minimumReleaseAge": "3 days",
+  "packageRules": [
+    {
+      "description": "Group all ExtraToast shared artifacts so a coherent platform bump lands in one PR.",
+      "matchPackagePatterns": ["^dev\\.extratoast", "^@extratoast/"],
+      "matchDepNames": ["ExtraToast/github-workflows"],
+      "groupName": "extratoast shared artifacts",
+      "labels": ["dependencies", "extratoast-platform"]
+    },
+    {
+      "description": "Pin GitHub Actions to immutable refs.",
+      "matchManagers": ["github-actions"],
+      "pinDigests": true
+    }
+  ],
+  "lockFileMaintenance": { "enabled": true }
+}