fix(ci): deploy-vps recreates PM2 via bin + gates on /login 200 (#3270)

Synchronized deploy-vps hardening (PM2 recreate via bin + /api/monitoring/health gate + fail-on-unhealthy). Supersedes #3262.

Integrated into release/v3.8.12.

.github/workflows/deploy-vps.yml

@@ -19,25 +19,54 @@ jobs:
     steps:
       - name: Deploy via SSH
         uses: appleboy/ssh-action@v1
-        continue-on-error: true
         with:
           host: ${{ secrets.VPS_HOST }}
           username: ${{ secrets.VPS_USER }}
           key: ${{ secrets.VPS_SSH_KEY }}
           port: 22
-          timeout: 30s
-          command_timeout: 5m
+          timeout: 60s
+          command_timeout: 15m
           script: |
-            echo "=== Updating OmniRoute ==="
-            npm install -g omniroute@latest 2>&1
-            INSTALLED_VERSION=$(omniroute --version 2>/dev/null || echo "unknown")
-            echo "Installed version: $INSTALLED_VERSION"
+            set -euo pipefail
 
-            echo "=== Restarting PM2 ==="
-            pm2 restart omniroute || pm2 start omniroute --name omniroute -- --port 20128
+            echo "=== Updating OmniRoute ==="
+            npm install -g omniroute@latest
+            INSTALLED_VERSION=$(omniroute --version 2>/dev/null | tr -d '[:space:]' || echo "unknown")
+            echo "Installed CLI version: $INSTALLED_VERSION"
+
+            # Recreate the PM2 process instead of `pm2 restart`. A bare restart
+            # re-runs whatever script path was saved earlier; after the build-output
+            # reorg (app/ -> dist/, .next -> .build/next) a process pinned to the old
+            # app/server-ws.mjs path can no longer start, and the node process dies
+            # while PM2 still reports "online" — so the box never binds :20128.
+            # Always launch via the `omniroute` bin so .env is loaded and the dist/
+            # layout is resolved correctly.
+            echo "=== (Re)creating PM2 process via bin ==="
+            pm2 delete omniroute 2>/dev/null || true
+            pm2 start omniroute --name omniroute -- --port 20128
             pm2 save
 
-            echo "=== Health Check ==="
-            sleep 3
-            curl -sf http://localhost:20128/api/settings > /dev/null && echo "✅ OmniRoute is healthy" || echo "❌ Health check failed"
+            # Health gate: fail the deploy unless the box actually reports healthy.
+            # Poll /api/monitoring/health for "status":"healthy" (a deeper signal than
+            # a static page 200 — it confirms the app booted, not just that a port is
+            # bound). Boot can take a while after a native-module/build-layout change,
+            # so poll up to ~3min before giving up.
+            echo "=== Health Check (gates the deploy) ==="
+            ok=0
+            for i in $(seq 1 36); do
+              BODY=$(curl -sf -m 5 http://localhost:20128/api/monitoring/health 2>/dev/null || true)
+              if printf '%s' "$BODY" | grep -q '"status":"healthy"'; then
+                ok=1
+                echo "✅ /api/monitoring/health -> healthy (attempt $i) — version $INSTALLED_VERSION"
+                break
+              fi
+              echo "… not healthy yet (attempt $i/36), retrying in 5s"
+              sleep 5
+            done
+            if [ "$ok" != "1" ]; then
+              echo "❌ Health check failed — /api/monitoring/health never reported healthy after ~3min"
+              echo "--- recent PM2 logs ---"
+              pm2 logs omniroute --lines 40 --nostream || true
+              exit 1
+            fi
             echo "=== Deploy complete ==="