Enable auth-service/api-gateway on production, build their images in CI

2026-08-18 14:11:00 +07:00
parent e5afedfa2f
commit b68005be1c
70 changed files with 6781 additions and 263 deletions
@@ -0,0 +1,85 @@
+name: Rollback k3s production
+
+# One-command escape hatch for medical-chatbot-app (ArgoCD, k3s) — the
+# Application build-practice-images.yml normally advances, and the same one
+# serving realvuxbaro.me since the 2026-08-17 cutover. This workflow does not
+# build anything: it only repoints the Application at an OLDER image tag that
+# a previous build-practice-images.yml run already pushed to GHCR, using the
+# exact same sync_practice_argocd.py logic that workflow uses to move
+# forward — a rollback is just that script pointed backward.
+#
+# Does not touch the Compose EC2 (stopped 2026-08-18, no CI/CD path left —
+# see docs/operations.md).
+
+on:
+  workflow_dispatch:
+    inputs:
+      target_sha:
+        description: >-
+          Commit SHA to roll back to. Must have a successful "Build and sync
+          k3s images" run (check: gh run list --workflow=build-practice-images.yml).
+        required: true
+
+concurrency:
+  # Same group as build-practice-images.yml: a rollback and a forward deploy
+  # must never race to repoint the same Application at the same time.
+  group: practice-images
+  cancel-in-progress: false
+
+jobs:
+  rollback:
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      packages: read
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: docker/setup-buildx-action@v3
+
+      - uses: docker/login-action@v3
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      # Fails fast and clearly on the most likely operator mistake: a typo'd
+      # or never-built SHA, rather than that surfacing later as a confusing
+      # ArgoCD/pod-level image pull failure.
+      - name: Confirm images exist for target_sha
+        run: |
+          set -eu
+          docker buildx imagetools inspect "ghcr.io/baovu2k4/vsf-duocthu-ai-service:${{ inputs.target_sha }}" > /dev/null
+          docker buildx imagetools inspect "ghcr.io/baovu2k4/vsf-duocthu-web:${{ inputs.target_sha }}" > /dev/null
+
+      - name: Point medical-chatbot-app at target_sha
+        env:
+          ARGOCD_PRACTICE_URL: ${{ secrets.ARGOCD_PRACTICE_URL }}
+          ARGOCD_PRACTICE_PASSWORD: ${{ secrets.ARGOCD_PRACTICE_PASSWORD }}
+          IMAGE_TAG: ${{ inputs.target_sha }}
+        run: python3 .github/scripts/sync_practice_argocd.py
+
+      # Blocks until the rollout has actually landed, not just until the sync
+      # call returned — sync_practice_argocd.py deliberately doesn't wait
+      # (see its own comment on the selfHeal race), so a rollback specifically
+      # needs this extra confirmation before it can claim success.
+      - name: Wait for the Application to be Synced, Healthy, and on target_sha
+        env:
+          ARGOCD_PRACTICE_URL: ${{ secrets.ARGOCD_PRACTICE_URL }}
+          ARGOCD_PRACTICE_PASSWORD: ${{ secrets.ARGOCD_PRACTICE_PASSWORD }}
+          EXPECT_TAG: ${{ inputs.target_sha }}
+        run: python3 .github/scripts/wait_for_argocd_sync.py
+
+      - name: Confirm realvuxbaro.me is serving the rolled-back build
+        run: |
+          for attempt in $(seq 1 18); do
+            code=$(curl -s -o /dev/null -w '%{http_code}' \
+              "https://realvuxbaro.me/api/history?conversation_id=rollback-smoke-${{ inputs.target_sha }}")
+            if [ "$code" = "200" ]; then
+              echo "realvuxbaro.me is live on ${{ inputs.target_sha }}"
+              exit 0
+            fi
+            sleep 10
+          done
+          echo "realvuxbaro.me did not respond healthy within 3 minutes after rollback"
+          exit 1