ops: make production builds atomic

scripts/deploy-production-atomic.sh

@@ -0,0 +1,52 @@
+#!/bin/sh
+set -eu
+
+app_dir="/srv/ai-chatbot"
+compose_dir="/srv/ai-chatbot-deploy"
+candidate_dir="$(mktemp -d "${app_dir}/.next-candidate.XXXXXX")"
+previous_dir="${app_dir}/.next-previous-$(date +%Y%m%dT%H%M%S)"
+build_container="ai-chatbot-build-$(date +%s)"
+
+cleanup_candidate() {
+  if [ -d "$candidate_dir" ]; then
+    rm -rf -- "$candidate_dir"
+  fi
+}
+
+trap cleanup_candidate EXIT INT TERM
+
+docker run --rm \
+  --name "$build_container" \
+  --network host \
+  -e NODE_OPTIONS=--max-old-space-size=6144 \
+  -v "${app_dir}:/workspace" \
+  -v "${candidate_dir}:/workspace/.next" \
+  -w /workspace \
+  node:20.20.0-bookworm-slim \
+  sh -lc 'corepack enable && pnpm build'
+
+test -s "${candidate_dir}/BUILD_ID"
+
+docker stop ai-chatbot-app >/dev/null
+if [ -d "${app_dir}/.next" ]; then
+  mv "${app_dir}/.next" "$previous_dir"
+fi
+mv "$candidate_dir" "${app_dir}/.next"
+candidate_dir=""
+
+docker compose -f "${compose_dir}/docker-compose.yml" up -d app
+
+if ! curl -fsS --retry 12 --retry-delay 2 --max-time 10 \
+  http://127.0.0.1:3125/login >/dev/null; then
+  docker stop ai-chatbot-app >/dev/null 2>&1 || true
+  failed_dir="${app_dir}/.next-failed-$(date +%Y%m%dT%H%M%S)"
+  mv "${app_dir}/.next" "$failed_dir"
+  if [ -d "$previous_dir" ]; then
+    mv "$previous_dir" "${app_dir}/.next"
+    docker compose -f "${compose_dir}/docker-compose.yml" up -d app
+  fi
+  echo "Deployment health check failed; previous build restored." >&2
+  exit 1
+fi
+
+echo "Production build activated. Previous build: ${previous_dir}"