baiyue-firstleg/deploy/scripts/deploy-built-service.sh

#!/usr/bin/env sh
set -eu

if [ "$#" -ne 5 ]; then
  echo "用法: $0 <环境> <镜像仓库> <镜像标签> <服务> <环境文件>" >&2
  exit 2
fi

DEPLOY_ENV="$1"
IMAGE_REPOSITORY="$2"
IMAGE_TAG="$3"
SERVICE="$4"
ENV_FILE="$5"
COMPOSE_FILE="deploy/docker-compose.yml"
IMAGE="$IMAGE_REPOSITORY/$SERVICE:$IMAGE_TAG"
REGISTRY_HOST="${IMAGE_REPOSITORY%%/*}"
PROJECT_NAME="firstleg-$DEPLOY_ENV"
HEALTH_TIMEOUT_SECONDS="${DEPLOY_HEALTH_TIMEOUT_SECONDS:-240}"
HEALTH_POLL_SECONDS="${DEPLOY_HEALTH_POLL_SECONDS:-5}"

for value_name in HEALTH_TIMEOUT_SECONDS HEALTH_POLL_SECONDS; do
  eval "value=\${$value_name}"
  case "$value" in
    ''|*[!0-9]*|0)
      echo "$value_name 必须为正整数，当前: $value" >&2
      exit 2
      ;;
  esac
done


case "$DEPLOY_ENV" in test|production) ;; *) echo "Only test and production are supported" >&2; exit 2 ;; esac
case "$SERVICE" in
  gateway|auth|iam|firstleg|finance|order|warehouse|file|print|integration-service|lastleg-service|warehouse-overseas-service|audit-service) ;;
  *) echo "Unknown or retired service: $SERVICE" >&2; exit 2 ;;
esac
case "$IMAGE_REPOSITORY" in ''|*[!a-zA-Z0-9._:/-]*) echo "Invalid image repository" >&2; exit 2 ;; esac
case "$IMAGE_TAG" in ''|*[!a-zA-Z0-9_.-]*) echo "Invalid image tag" >&2; exit 2 ;; esac
case "${ALLOW_SAME_CONTRACT_ROLLBACK:-false}" in true|false) ;; *) exit 2 ;; esac
if [ "$DEPLOY_ENV" = "test" ] && [ "${TARGET_IS_LOCAL:-false}" != "true" ]; then
  echo "TEST deploy must run on the target host" >&2; exit 2
fi
[ -f "$ENV_FILE" ] || { echo "Environment file missing" >&2; exit 2; }
env_file_value() { sed -n "s/^$1=//p" "$ENV_FILE" | tail -n 1 | tr -d '\r'; }
for key in NACOS_SERVER_ADDR NACOS_NAMESPACE NACOS_USERNAME NACOS_PASSWORD REDIS_HOST REDIS_PORT REDIS_DATABASE REDIS_TOKEN_KEY_PREFIX INTERNAL_SERVICE_TOKEN JWT_RSA_JWK_SET JWT_CURRENT_KID ROCKETMQ_NAME_SERVER; do
  [ -n "$(env_file_value "$key")" ] || { echo "Missing $key" >&2; exit 2; }
done
[ "$(env_file_value MQ_TOPIC_PREFIX)" = "$DEPLOY_ENV-" ] || { echo "MQ_TOPIC_PREFIX must be $DEPLOY_ENV-" >&2; exit 2; }
case "$SERVICE" in
  gateway|print|integration-service) db_prefix='' ;;
  lastleg-service) db_prefix=LASTLEG ;;
  warehouse-overseas-service) db_prefix=WAREHOUSE_OVERSEAS ;;
  audit-service) db_prefix=AUDIT ;;
  *) db_prefix=$(printf '%s' "$SERVICE" | tr '[:lower:]' '[:upper:]') ;;
esac
if [ -n "$db_prefix" ]; then
  for suffix in URL USERNAME PASSWORD; do
    key="${db_prefix}_DB_${suffix}"
    [ -n "$(env_file_value "$key")" ] || { echo "Missing $key" >&2; exit 2; }
  done
fi
if [ "$SERVICE" = "firstleg" ] && [ "$(env_file_value TERMINAL_DRAYAGE_MAIL_ENABLED)" = "true" ]; then
  for key in TERMINAL_DRAYAGE_MAIL_HOST TERMINAL_DRAYAGE_MAIL_PORT TERMINAL_DRAYAGE_MAIL_USERNAME TERMINAL_DRAYAGE_MAIL_PASSWORD; do
    [ -n "$(env_file_value "$key")" ] || { echo "Missing $key" >&2; exit 2; }
  done
fi
# Architecture/schema changes must use whole-release recovery, never an implicit old image.
export PUBLISHED_BIND_ADDRESS=127.0.0.1
: "${REGISTRY_USER:?REGISTRY_USER 未设置}"
: "${REGISTRY_PASSWORD:?REGISTRY_PASSWORD 未设置}"

local_registry_logout() {
  docker logout "$REGISTRY_HOST" >/dev/null 2>&1 || true
  if [ -n "${deploy_docker_config:-}" ]; then rm -rf -- "$deploy_docker_config"; fi
}

local_compose() {
  docker compose \
    --project-name "$PROJECT_NAME" \
    --env-file "$ENV_FILE" \
    -f "$COMPOSE_FILE" \
    "$@"
}

local_container_id() {
  local_compose ps --all --quiet "$SERVICE" | head -n 1
}

local_container_image() {
  container_id="$(local_container_id)"
  if [ -n "$container_id" ]; then
    docker inspect --format '{{.Config.Image}}' "$container_id"
  fi
}

local_health_state() {
  container_id="$(local_container_id)"
  if [ -z "$container_id" ]; then
    echo missing
    return
  fi
  docker inspect --format '{{if .State.Health}}{{.State.Health.Status}}{{else}}no-health:{{.State.Status}}{{end}}' "$container_id" 2>/dev/null || echo missing
}

local_runtime_state() {
  container_id="$(local_container_id)"
  if [ -z "$container_id" ]; then
    echo missing
    return
  fi
  docker inspect --format '{{.State.Status}}' "$container_id" 2>/dev/null || echo missing
}

wait_for_local_health() {
  elapsed=0
  while [ "$elapsed" -lt "$HEALTH_TIMEOUT_SECONDS" ]; do
    runtime_state="$(local_runtime_state)"
    health_state="$(local_health_state)"
    echo "容器运行状态: $runtime_state，健康状态: $health_state（已等待 ${elapsed}s）"
    case "$runtime_state" in
      restarting|exited|dead)
        echo "容器未保持运行，立即停止等待: $runtime_state" >&2
        return 1
        ;;
    esac
    case "$health_state" in
      healthy)
        return 0
        ;;
      no-health:running)
        echo "Missing HEALTHCHECK; deployment failed" >&2
        return 1
        ;;
      unhealthy|no-health:exited|no-health:dead)
        return 1
        ;;
    esac
    sleep "$HEALTH_POLL_SECONDS"
    elapsed=$((elapsed + HEALTH_POLL_SECONDS))
  done
  echo "等待容器健康超时: ${HEALTH_TIMEOUT_SECONDS}s" >&2
  return 1
}

deploy_local() {
  export DEPLOY_ENV IMAGE_REPOSITORY IMAGE_TAG
  deploy_docker_config=$(mktemp -d)
  DOCKER_CONFIG="$deploy_docker_config"
  export DOCKER_CONFIG
  trap local_registry_logout EXIT HUP INT TERM
  printf '%s' "$REGISTRY_PASSWORD" | docker login -u "$REGISTRY_USER" --password-stdin "$REGISTRY_HOST"
  trap local_registry_logout EXIT HUP INT TERM
  previous_image="$(local_container_image || true)"
  if [ -n "$previous_image" ]; then
    echo "部署前运行镜像: $previous_image"
  else
    echo "首次部署，未发现可回滚的旧镜像"
  fi
  docker pull "$IMAGE"
  export DEPLOY_ENV IMAGE_REPOSITORY IMAGE_TAG
  new_started=true
  if ! local_compose up -d --no-deps --pull never "$SERVICE"; then
    echo "新镜像启动命令失败: $IMAGE" >&2
    new_started=false
  fi
  if [ "$new_started" = "true" ] && wait_for_local_health; then
    running_image="$(local_container_image || true)"
    if [ "$running_image" != "$IMAGE" ]; then
      echo "运行容器镜像与请求镜像不一致: expected=$IMAGE actual=${running_image:-missing}" >&2
      return 1
    fi
    echo "新镜像健康检查通过: $IMAGE"
    local_compose ps "$SERVICE"
    local_registry_logout
    trap - EXIT HUP INT TERM
    return 0
  fi

  echo "新镜像健康检查失败: $IMAGE" >&2
  failed_container_id="$(local_container_id || true)"
  if [ -n "$failed_container_id" ]; then
    echo "新镜像容器最近日志（最多 200 行）:" >&2
    docker logs --tail 200 "$failed_container_id" 2>&1 || true
  else
    echo "新镜像容器不存在，无法读取启动日志" >&2
  fi
  if [ "${ALLOW_SAME_CONTRACT_ROLLBACK:-false}" != "true" ]; then
    echo "Automatic rollback disabled: keep writes closed and recover the matched release" >&2
    local_compose stop "$SERVICE" >/dev/null 2>&1 || true
    return 1
  fi
  expected_prefix="$IMAGE_REPOSITORY/$SERVICE:"
  case "$previous_image" in
    "$expected_prefix"*)
      previous_tag="${previous_image#"$expected_prefix"}"
      ;;
    '')
      echo "首次部署没有可回滚镜像，部署失败" >&2
      local_compose stop "$SERVICE" >/dev/null 2>&1 || true
      return 1
      ;;
    *)
      echo "旧镜像不属于当前服务，拒绝自动回滚: $previous_image" >&2
      return 1
      ;;
  esac
  if [ "$previous_image" = "$IMAGE" ]; then
    echo "旧镜像与新镜像相同，无法自动回滚: $IMAGE" >&2
    return 1
  fi

  echo "开始自动回滚到: $previous_image" >&2
  if ! docker pull "$previous_image"; then
    echo "自动回滚失败，无法拉取旧镜像: $previous_image" >&2
    return 1
  fi
  IMAGE_TAG="$previous_tag"
  export IMAGE_TAG
  if ! local_compose up -d --no-deps --pull never "$SERVICE"; then
    echo "自动回滚失败，旧镜像启动命令失败: $previous_image" >&2
    return 1
  fi
  if wait_for_local_health; then
    echo "自动回滚成功: $previous_image" >&2
    local_compose ps "$SERVICE" || true
  else
    echo "自动回滚失败，旧镜像未恢复健康: $previous_image" >&2
  fi
  return 1
}

deploy_remote() {
  : "${TARGET_SSH_TARGET:?TARGET_SSH_TARGET required}"
  : "${TARGET_SSH_KEY:?TARGET_SSH_KEY required}"
  : "${TARGET_SSH_KNOWN_HOSTS:?TARGET_SSH_KNOWN_HOSTS required}"
  [ -f "$TARGET_SSH_KEY" ] && [ -f "$TARGET_SSH_KNOWN_HOSTS" ] || exit 2
  run_ssh() {
    ssh -i "$TARGET_SSH_KEY" -o BatchMode=yes -o StrictHostKeyChecking=yes \
      -o UserKnownHostsFile="$TARGET_SSH_KNOWN_HOSTS" -o ConnectTimeout=10 "$TARGET_SSH_TARGET" "$@"
  }
  run_scp() {
    scp -i "$TARGET_SSH_KEY" -o BatchMode=yes -o StrictHostKeyChecking=yes \
      -o UserKnownHostsFile="$TARGET_SSH_KNOWN_HOSTS" -o ConnectTimeout=10 "$@"
  }
  remote_dir=$(run_ssh 'umask 077; mktemp -d /tmp/firstleg-production-release.XXXXXXXXXX')
  case "$remote_dir" in /tmp/firstleg-production-release.*) ;; *) echo "Invalid remote temporary directory" >&2; exit 2 ;; esac
  case "$remote_dir" in *[!a-zA-Z0-9/._-]*) exit 2 ;; esac
  cleanup_remote() { run_ssh "rm -rf -- '$remote_dir'" >/dev/null 2>&1; }
  trap cleanup_remote EXIT HUP INT TERM
  run_ssh "mkdir -p '$remote_dir/deploy/scripts'"
  run_scp "$ENV_FILE" "$TARGET_SSH_TARGET:$remote_dir/deployment.env"
  run_scp "$COMPOSE_FILE" "$TARGET_SSH_TARGET:$remote_dir/deploy/docker-compose.yml"
  run_scp deploy/scripts/deploy-built-service.sh "$TARGET_SSH_TARGET:$remote_dir/deploy/scripts/deploy-built-service.sh"
  # Registry credentials only travel over stdin, never in process arguments or files.
  { printf '%s\n' "$REGISTRY_USER"; printf '%s' "$REGISTRY_PASSWORD"; } | run_ssh \
    "cd '$remote_dir' && IFS= read -r REGISTRY_USER && REGISTRY_PASSWORD=\$(cat) && export REGISTRY_USER REGISTRY_PASSWORD TARGET_IS_LOCAL=true ALLOW_SAME_CONTRACT_ROLLBACK='${ALLOW_SAME_CONTRACT_ROLLBACK:-false}' DEPLOY_HEALTH_TIMEOUT_SECONDS='$HEALTH_TIMEOUT_SECONDS' DEPLOY_HEALTH_POLL_SECONDS='$HEALTH_POLL_SECONDS' && sh deploy/scripts/deploy-built-service.sh production '$IMAGE_REPOSITORY' '$IMAGE_TAG' '$SERVICE' deployment.env"
  cleanup_remote
  trap - EXIT HUP INT TERM
}

if [ "${TARGET_IS_LOCAL:-false}" = "true" ]; then
  deploy_local
else
  deploy_remote
fi