fix prod deployment

production/deploy-production.sh

@@ -0,0 +1,326 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+cd "${ROOT_DIR}"
+
+source "${ROOT_DIR}/scripts/docker-prod-common.sh"
+
+usage() {
+  cat <<'EOF'
+Usage:
+  ./deploy-production.sh [--with-pgmanage] [--skip-prune]
+
+What it does:
+  1. Validates .env.docker.production
+  2. Loads IMAGE_TAG, APP_IMAGE, registry, and archive settings from the env file
+  3. Loads a matching Docker image archive from IMAGE_ARCHIVE_DIR when present
+  4. Otherwise logs in to the registry if credentials are configured, then pulls
+  5. Reuses healthy Traefik, Postgres, and Redis containers when present
+  6. Waits for health checks and prints container status
+
+VPN/shared-folder image flow:
+  Put an archive matching IMAGE_TAG in ./image-archives before running this script.
+  Example: ./image-archives/carmanagement-latest.tar
+EOF
+}
+
+include_pgmanage=0
+skip_prune=0
+
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    --with-pgmanage)
+      include_pgmanage=1
+      shift
+      ;;
+    --skip-prune)
+      skip_prune=1
+      shift
+      ;;
+    -h|--help)
+      usage
+      exit 0
+      ;;
+    *)
+      echo "Unknown option: $1" >&2
+      usage >&2
+      exit 1
+      ;;
+  esac
+done
+
+require_command() {
+  local command_name="$1"
+
+  if ! command -v "${command_name}" >/dev/null 2>&1; then
+    echo "Missing required command: ${command_name}" >&2
+    exit 1
+  fi
+}
+
+normalize_archive_dir() {
+  if [[ -z "${IMAGE_ARCHIVE_DIR:-}" ]]; then
+    return 0
+  fi
+
+  case "${IMAGE_ARCHIVE_DIR}" in
+    /*)
+      ;;
+    *)
+      IMAGE_ARCHIVE_DIR="${ROOT_DIR}/${IMAGE_ARCHIVE_DIR#./}"
+      export IMAGE_ARCHIVE_DIR
+      ;;
+  esac
+}
+
+print_release_summary() {
+  echo "Production deployment"
+  echo "  Image: ${APP_IMAGE:-rentaldrivego/carmanagement}:${IMAGE_TAG}"
+  echo "  Env: ${ENV_FILE}"
+  echo "  Compose project: ${PROJECT_NAME}"
+
+  if [[ -n "${IMAGE_ARCHIVE_DIR:-}" ]]; then
+    echo "  Image archive dir: ${IMAGE_ARCHIVE_DIR}"
+  fi
+}
+
+container_status() {
+  local container_name="$1"
+
+  docker inspect --format='{{if .State.Health}}{{.State.Health.Status}}{{else}}{{.State.Status}}{{end}}' "${container_name}" 2>/dev/null || echo "missing"
+}
+
+compose_service_status() {
+  local compose_name="$1"
+  local service_name="$2"
+  local container_id
+
+  case "${compose_name}" in
+    prod)
+      container_id="$(prod_compose ps -q "${service_name}" 2>/dev/null || true)"
+      ;;
+    traefik)
+      container_id="$(traefik_compose ps -q "${service_name}" 2>/dev/null || true)"
+      ;;
+    *)
+      echo "missing"
+      return 0
+      ;;
+  esac
+
+  if [[ -z "${container_id}" ]]; then
+    echo "missing"
+    return 0
+  fi
+
+  docker inspect --format='{{if .State.Health}}{{.State.Health.Status}}{{else}}{{.State.Status}}{{end}}' "${container_id}" 2>/dev/null || echo "missing"
+}
+
+first_container_id() {
+  docker ps -q "$@" 2>/dev/null | sed -n '1p'
+}
+
+ensure_container_on_traefik_network() {
+  local container_id="$1"
+  local networks
+
+  networks="$(docker inspect --format='{{range $name, $_ := .NetworkSettings.Networks}}{{printf "%s " $name}}{{end}}' "${container_id}" 2>/dev/null || true)"
+  case " ${networks} " in
+    *" ${TRAEFIK_NETWORK} "*)
+      ;;
+    *)
+      echo "Connecting existing Traefik container ${container_id} to ${TRAEFIK_NETWORK}."
+      docker network connect "${TRAEFIK_NETWORK}" "${container_id}" 2>/dev/null || true
+      ;;
+  esac
+}
+
+find_running_traefik_container() {
+  local container_id
+
+  container_id="$(first_container_id --filter name=traefik)"
+  if [[ -n "${container_id}" ]]; then
+    printf '%s' "${container_id}"
+    return 0
+  fi
+
+  container_id="$(first_container_id --filter ancestor=traefik --filter status=running)"
+  if [[ -n "${container_id}" ]]; then
+    printf '%s' "${container_id}"
+    return 0
+  fi
+
+  container_id="$(first_container_id --filter ancestor=traefik:latest --filter status=running)"
+  if [[ -n "${container_id}" ]]; then
+    printf '%s' "${container_id}"
+    return 0
+  fi
+}
+
+ensure_ports_available_for_traefik() {
+  local container_id
+  local port_listener
+
+  container_id="$(first_container_id --filter publish=80 --filter status=running)"
+  if [[ -n "${container_id}" ]]; then
+    echo "Port 80 is already used by container ${container_id}, and it does not look like Traefik." >&2
+    echo "Stop that container or set up Traefik outside this deployment before rerunning." >&2
+    exit 1
+  fi
+
+  port_listener="$(sudo ss -H -ltnp 'sport = :80' 2>/dev/null || ss -H -ltnp 'sport = :80' 2>/dev/null || true)"
+  if [[ -n "${port_listener}" ]]; then
+    echo "Port 80 is already used by a host process:" >&2
+    echo "${port_listener}" >&2
+    echo "Stop that process before starting Traefik, or manage Traefik outside this deployment." >&2
+    exit 1
+  fi
+
+  container_id="$(first_container_id --filter publish=443 --filter status=running)"
+  if [[ -n "${container_id}" ]]; then
+    echo "Port 443 is already used by container ${container_id}, and it does not look like Traefik." >&2
+    echo "Stop that container or set up Traefik outside this deployment before rerunning." >&2
+    exit 1
+  fi
+
+  port_listener="$(sudo ss -H -ltnp 'sport = :443' 2>/dev/null || ss -H -ltnp 'sport = :443' 2>/dev/null || true)"
+  if [[ -n "${port_listener}" ]]; then
+    echo "Port 443 is already used by a host process:" >&2
+    echo "${port_listener}" >&2
+    echo "Stop that process before starting Traefik, or manage Traefik outside this deployment." >&2
+    exit 1
+  fi
+}
+
+ensure_traefik_ready() {
+  local existing_container_id
+  local status
+
+  status="$(compose_service_status traefik traefik)"
+  case "${status}" in
+    running|healthy)
+      echo "Traefik is already ${status}; reusing existing container."
+      ;;
+    missing|exited|dead|created|restarting|removing|paused|unhealthy)
+      existing_container_id="$(find_running_traefik_container)"
+      if [[ -n "${existing_container_id}" ]]; then
+        echo "Found existing Traefik container ${existing_container_id}; reusing it."
+        ensure_container_on_traefik_network "${existing_container_id}"
+        return 0
+      fi
+      ensure_ports_available_for_traefik
+      echo "Starting Traefik (status: ${status})"
+      start_traefik
+      ;;
+    *)
+      existing_container_id="$(find_running_traefik_container)"
+      if [[ -n "${existing_container_id}" ]]; then
+        echo "Found existing Traefik container ${existing_container_id}; reusing it."
+        ensure_container_on_traefik_network "${existing_container_id}"
+        return 0
+      fi
+      ensure_ports_available_for_traefik
+      echo "Starting Traefik (status: ${status})"
+      start_traefik
+      ;;
+  esac
+}
+
+ensure_postgres_ready() {
+  local status
+
+  status="$(compose_service_status prod postgres)"
+  case "${status}" in
+    healthy)
+      echo "Postgres is already healthy; reusing existing container."
+      ;;
+    running)
+      echo "Postgres is running; waiting for health check."
+      wait_for_healthy postgres 120
+      ;;
+    missing|exited|dead|created|restarting|removing|paused|unhealthy)
+      echo "Starting Postgres (status: ${status})"
+      prod_compose up -d postgres
+      wait_for_healthy postgres 120
+      ;;
+    *)
+      echo "Starting Postgres (status: ${status})"
+      prod_compose up -d postgres
+      wait_for_healthy postgres 120
+      ;;
+  esac
+}
+
+ensure_redis_ready() {
+  local status
+
+  status="$(compose_service_status prod redis)"
+  case "${status}" in
+    healthy)
+      echo "Redis is already healthy; reusing existing container."
+      ;;
+    running)
+      echo "Redis is running; waiting for health check."
+      wait_for_healthy redis 120
+      ;;
+    missing|exited|dead|created|restarting|removing|paused|unhealthy)
+      echo "Starting Redis (status: ${status})"
+      prod_compose up -d redis
+      wait_for_healthy redis 120
+      ;;
+    *)
+      echo "Starting Redis (status: ${status})"
+      prod_compose up -d redis
+      wait_for_healthy redis 120
+      ;;
+  esac
+}
+
+require_command docker
+
+ensure_env_file
+load_prod_env_exports
+require_release_image
+normalize_archive_dir
+print_release_summary
+
+ensure_traefik_network
+ensure_api_uploads_volume
+login_registry_if_configured
+
+echo "Preparing release image"
+pull_prod_release_images
+
+echo "Checking shared infrastructure"
+ensure_traefik_ready
+ensure_postgres_ready
+ensure_redis_ready
+
+echo "Running database migrations"
+run_prod_migrations
+
+echo "Starting application services"
+app_services=(api storefront dashboard admin)
+
+if [[ "${include_pgmanage}" -eq 1 ]]; then
+  prod_compose --profile private-tools up -d "${app_services[@]}" pgmanage
+else
+  prod_compose up -d "${app_services[@]}"
+fi
+
+wait_for_healthy api 180
+wait_for_healthy storefront 180
+wait_for_healthy dashboard 180
+wait_for_healthy admin 180
+
+if [[ "${skip_prune}" -eq 0 ]]; then
+  echo "Pruning dangling images"
+  docker image prune -f >/dev/null
+fi
+
+echo "Final container status"
+prod_compose ps
+
+echo "Production release ${APP_IMAGE:-rentaldrivego/carmanagement}:${IMAGE_TAG} is up and healthy."