fix health check + build

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

@@ -12,16 +12,25 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v3
 
-      - name: Build Docker Image
+      # The Dockerfile has two runnable stages and no sensible default: an
+      # untargeted build produces the last one (frontend) on its own, which
+      # then has nothing to proxy /api to. Both get published separately.
+      - name: Build Docker Images
         run: |
-          docker build -t git.alexav.gg/${{ gitea.repository }}:latest .
-          docker build -t git.alexav.gg/${{ gitea.repository }}:${{ gitea.sha }} .
+          docker build --target backend \
+            -t git.alexav.gg/${{ gitea.repository }}:backend-latest \
+            -t git.alexav.gg/${{ gitea.repository }}:backend-${{ gitea.sha }} .
+          docker build --target frontend \
+            -t git.alexav.gg/${{ gitea.repository }}:frontend-latest \
+            -t git.alexav.gg/${{ gitea.repository }}:frontend-${{ gitea.sha }} .
 
       - name: Login to Gitea Registry
         run: |
           echo "${{ secrets.RUNNER_TOKEN }}" | docker login git.alexav.gg -u ${{ gitea.actor }} --password-stdin
 
-      - name: Push Docker Image
+      - name: Push Docker Images
         run: |
-          docker push git.alexav.gg/${{ gitea.repository }}:latest
-          docker push git.alexav.gg/${{ gitea.repository }}:${{ gitea.sha }}
+          docker push git.alexav.gg/${{ gitea.repository }}:backend-latest
+          docker push git.alexav.gg/${{ gitea.repository }}:backend-${{ gitea.sha }}
+          docker push git.alexav.gg/${{ gitea.repository }}:frontend-latest
+          docker push git.alexav.gg/${{ gitea.repository }}:frontend-${{ gitea.sha }}

Dockerfile

@@ -99,7 +99,9 @@ USER node
 WORKDIR /app/apps/frontend
 
 EXPOSE 5173
+# /healthz, not /: rendering / reaches the backend, so probing it would report
+# a backend outage as a dead frontend, and Traefik drops unhealthy containers.
 HEALTHCHECK --interval=30s --timeout=5s --start-period=20s --retries=3 \
-  CMD ["node", "-e", "fetch('http://127.0.0.1:'+(process.env.PORT||5173)+'/').then(r=>process.exit(r.ok?0:1)).catch(()=>process.exit(1))"]
+  CMD ["node", "-e", "fetch('http://127.0.0.1:'+(process.env.PORT||5173)+'/healthz').then(r=>process.exit(r.ok?0:1)).catch(()=>process.exit(1))"]
 
 CMD ["node", "server.js", "--prod"]

apps/frontend/server.js

@@ -12,6 +12,11 @@ const backendOrigin = process.env.BACKEND_ORIGIN ?? "http://localhost:3000";
 
 const app = express();
 
+// Liveness only, so it must not touch the backend: the container healthcheck
+// reads this, and an unhealthy container is dropped by Traefik's Docker
+// provider - which would take the router, and with it the certificate, down.
+app.get("/healthz", (req, res) => res.status(200).type("text/plain").end("ok"));
+
 /** @type {import("vite").ViteDevServer | undefined} */
 let vite;