fix: keep app online during deploy preparation

.gitea/workflows/ci.yaml

@@ -70,16 +70,19 @@ jobs:
           error_handler() {
             cd /var/www/atom-nexst 2>/dev/null || cd / || true
             echo "!!! DEPLOYMENT FAILED on line $1 !!!" >&2
-            # Roll back the build artifact if cutover already moved .next into place.
-            if [ "${CUTOVER_STARTED}" = "1" ] && [ -d "${LIVE}/.next.prev" ]; then
-              echo "Rolling back .next to previous artifact..." >&2
-              rm -rf "${LIVE}/.next" || true
-              mv "${LIVE}/.next.prev" "${LIVE}/.next" || true
+            # Leave the healthy current process untouched when staging fails.
+            # Restart only when cutover has already stopped or replaced it.
+            if [ "${CUTOVER_STARTED}" = "1" ]; then
+              if [ -d "${LIVE}/.next.prev" ]; then
+                echo "Rolling back .next to previous artifact..." >&2
+                rm -rf "${LIVE}/.next" || true
+                mv "${LIVE}/.next.prev" "${LIVE}/.next" || true
+              fi
+              pm2 restart next --update-env 2>/dev/null || pm2 start pnpm --name "next" -- start 2>/dev/null || true
             fi
             if [ -n "${STAGE}" ] && [ -d "${STAGE}" ]; then
               git -C "${LIVE}" worktree remove --force "${STAGE}" 2>/dev/null || rm -rf "${STAGE}" || true
             fi
-            pm2 restart next --update-env 2>/dev/null || pm2 start pnpm --name "next" -- start 2>/dev/null || true
             exit 1
           }
           trap 'error_handler $LINENO' ERR
@@ -150,14 +153,7 @@ jobs:
             exit 1
           fi
 
-          echo "Cutover: stop service (free DB connections), migrate, swap .next..."
-          CUTOVER_STARTED=1
-          pm2 stop next --kill-timeout 10000 || true
-          # Wait for PM2 to fully exit and MariaDB to reclaim connections.
-          sleep 10
-
-          # Migrate only after live is stopped — avoids ER_CON_COUNT_ERROR while
-          # the old process still holds DATABASE_POOL_SIZE connections.
+          echo "Migrating staged release while the current app stays online..."
           cd "${STAGE}"
           MIGRATE_OK=0
           for i in $(seq 1 10); do
@@ -173,6 +169,10 @@ jobs:
             exit 1
           fi
 
+          echo "Cutover: stop service and swap .next..."
+          CUTOVER_STARTED=1
+          pm2 stop next --kill-timeout 10000 || true
+
           cd "${LIVE}"
           echo "Hard reset live tree to origin/main (no nuclear src wipe)..."
           git reset --hard origin/main