vmis/backend/app/services/scheduler/schedule_tenant.py

"""
Schedule 1 — 租戶檢查（每 3 分鐘）
3-state: None=未設定(灰), True=正常(綠), False=異常(紅)
- None  → 自動嘗試建立/部署，記錄 done_at
- False → 發送告警 email 給所有管理員
"""
import logging
import smtplib
from email.mime.text import MIMEText
from typing import Optional

import httpx
from sqlalchemy.orm import Session

from app.core.config import settings
from app.core.utils import now_tw
from app.models.tenant import Tenant
from app.models.result import TenantScheduleResult

logger = logging.getLogger(__name__)

PG_PORT_TRIAL = 5433
PG_PORT_ACTIVE = 5432
PG_USER = "admin"
PG_PASS = "DC1qaz2wsx"
PG_HOST_TRIAL = "10.1.0.20"
PG_HOST_ACTIVE = "10.1.0.254"

REDIS_HOST_TRIAL = "10.14.0.20"
REDIS_HOST_ACTIVE = "10.1.0.254"
REDIS_PORT = 6379
REDIS_PASS = "DC1qaz2wsx"

KC_HOST_TRIAL = "auth.lab.taipei"
KC_HOST_ACTIVE = "auth.ease.taipei"

TRAEFIK_DYNAMIC_DIR = "/home/porsche/traefik/dynamic"
TRAEFIK_API_URL = "http://10.1.0.254:8080"


# ─── Docker Compose 範本產生 ─────────────────────────────────────────────────

OO_AI_PLUGIN_GUID = "{9DC93CDB-B576-4F0C-B55E-FCC9C48DD007}"
OO_AI_TRANSLATIONS_HOST = "/home/porsche/tenants/shared/oo-plugins/ai-translations"
OO_AI_TRANSLATIONS_CONTAINER = (
    f"/var/www/onlyoffice/documentserver/sdkjs-plugins/{OO_AI_PLUGIN_GUID}/translations"
)


def _generate_tenant_compose(tenant, is_active: bool) -> str:
    """
    產生租戶 docker-compose.yml 內容。
    包含 NC + OO 容器設定，OO 已加入繁中 AI plugin bind mount。
    """
    code = tenant.code
    suffix = "" if is_active else "-test"
    nc = f"nc-{code}{suffix}"
    oo = f"oo-{code}{suffix}"
    pg_host = PG_HOST_ACTIVE if is_active else PG_HOST_TRIAL
    pg_port = PG_PORT_ACTIVE if is_active else PG_PORT_TRIAL
    pg_db = f"nc_{code}_db"
    nc_domain = tenant.domain
    oo_host = f"office-{code}.ease.taipei" if is_active else f"office-{code}.lab.taipei"

    ai_base = OO_AI_TRANSLATIONS_HOST
    ai_cont = OO_AI_TRANSLATIONS_CONTAINER

    return f"""services:
  {nc}:
    image: nextcloud:31
    container_name: {nc}
    restart: unless-stopped
    volumes:
      - {nc}-data:/var/www/html
      - {nc}-apps:/var/www/html/custom_apps
      - {nc}-config:/var/www/html/config
    environment:
      POSTGRES_HOST: {pg_host}:{pg_port}
      POSTGRES_DB: {pg_db}
      POSTGRES_USER: {PG_USER}
      POSTGRES_PASSWORD: ${{NC_DB_PASSWORD}}
      NEXTCLOUD_ADMIN_USER: ${{NC_ADMIN_USER}}
      NEXTCLOUD_ADMIN_PASSWORD: ${{NC_ADMIN_PASSWORD}}
      NEXTCLOUD_TRUSTED_DOMAINS: {nc_domain}
      OVERWRITEPROTOCOL: https
      TRUSTED_PROXIES: 172.18.0.0/16
      TZ: Asia/Taipei
    networks:
      - traefik-network
    labels:
      - "traefik.enable=false"

  {oo}:
    image: onlyoffice/documentserver:latest
    container_name: {oo}
    restart: unless-stopped
    environment:
      - JWT_SECRET=${{OO_JWT_SECRET}}
    volumes:
      - {oo}-data:/var/www/onlyoffice/Data
      - {oo}-log:/var/log/onlyoffice
      - {ai_base}/zh-TW.json:{ai_cont}/zh-TW.json:ro
      - {ai_base}/zh-TW.json.gz:{ai_cont}/zh-TW.json.gz:ro
      - {ai_base}/langs.json:{ai_cont}/langs.json:ro
      - {ai_base}/langs.json.gz:{ai_cont}/langs.json.gz:ro
    networks:
      - traefik-network
    labels:
      - "traefik.enable=true"
      - "traefik.docker.network=traefik-network"
      - "traefik.http.routers.{oo}.rule=Host(`{oo_host}`)"
      - "traefik.http.routers.{oo}.entrypoints=websecure"
      - "traefik.http.routers.{oo}.tls=true"
      - "traefik.http.routers.{oo}.tls.certresolver=letsencrypt"
      - "traefik.http.services.{oo}.loadbalancer.server.port=80"
      - "traefik.http.middlewares.{oo}-headers.headers.customrequestheaders.X-Forwarded-Proto=https"
      - "traefik.http.routers.{oo}.middlewares={oo}-headers"

networks:
  traefik-network:
    external: true

volumes:
  {nc}-data:
  {nc}-apps:
  {nc}-config:
  {oo}-data:
  {oo}-log:
"""


def _ensure_tenant_compose(tenant, is_active: bool) -> bool:
    """
    確保租戶 docker-compose.yml 存在且為最新範本。
    若不存在則自動產生並寫入，同時補齊 .env。
    """
    try:
        import paramiko
        code = tenant.code
        deploy_name = code if is_active else f"{code}-test"
        deploy_dir = f"{settings.TENANT_DEPLOY_BASE}/{deploy_name}"
        compose_path = f"{deploy_dir}/docker-compose.yml"
        env_path = f"{deploy_dir}/.env"

        client = paramiko.SSHClient()
        client.set_missing_host_key_policy(paramiko.AutoAddPolicy())
        client.connect(settings.DOCKER_SSH_HOST, username=settings.DOCKER_SSH_USER, timeout=15)

        # 確保目錄存在
        _, out, _ = client.exec_command(f"mkdir -p {deploy_dir}")
        out.channel.recv_exit_status()

        # 若 docker-compose.yml 不存在才寫入（避免覆蓋手動調整）
        _, stdout, _ = client.exec_command(f"test -f {compose_path} && echo exists || echo missing")
        exists = stdout.read().decode().strip() == "exists"

        if not exists:
            content = _generate_tenant_compose(tenant, is_active)
            sftp = client.open_sftp()
            with sftp.open(compose_path, "w") as f:
                f.write(content)
            sftp.close()
            logger.info(f"docker-compose.yml generated for {code}")

        # 確保 .env 存在（含必要變數）
        _, stdout2, _ = client.exec_command(f"test -f {env_path} && echo exists || echo missing")
        env_exists = stdout2.read().decode().strip() == "exists"
        if not env_exists:
            from app.core.config import settings as _cfg
            env_content = (
                f"NC_DB_PASSWORD={PG_PASS}\n"
                f"NC_ADMIN_USER={_cfg.NC_ADMIN_USER}\n"
                f"NC_ADMIN_PASSWORD={_cfg.NC_ADMIN_PASSWORD}\n"
                f"OO_JWT_SECRET={_cfg.OO_JWT_SECRET}\n"
            )
            sftp2 = client.open_sftp()
            with sftp2.open(env_path, "w") as f:
                f.write(env_content)
            sftp2.close()
            logger.info(f".env generated for {code}")

        client.close()
        return True
    except Exception as e:
        logger.error(f"_ensure_tenant_compose {tenant.code}: {e}")
        return False


# ─── Traefik file provider helpers ───────────────────────────────────────────

def _generate_tenant_route_yaml(tenant, is_active: bool) -> str:
    """
    產生租戶 Traefik 路由 YAML 內容。
    租戶網域根路徑直接指向 NC（無路徑前綴）。
    is_manager=true 的租戶額外加入 /admin 路由（priority 200，比 drive 高）。
    """
    code = tenant.code
    domain = tenant.domain
    nc_url = f"http://nc-{code}:80" if is_active else f"http://nc-{code}-test:80"

    lines = ["http:"]

    if tenant.is_manager:
        # Manager 租戶：/admin → vmis-backend，/api → vmis-backend，根路徑 → NC Drive
        realm = tenant.code
        vmis_backend_url = f"http://{realm}:10281" if is_active else f"http://{realm}-test:10281"
        lines += [
            "  middlewares:",
            "    vmis-strip-admin:",
            "      stripPrefix:",
            '        prefixes: ["/admin"]',
            f"    {code}-oidc-redirect:",
            "      redirectRegex:",
            '        regex: ".*"',
            f'        replacement: "https://{domain}/apps/user_oidc/login/1"',
            "        permanent: false",
            "",
        ]
        lines += [
            "  routers:",
            f"    {code}-login:",
            f'      rule: "Host(`{domain}`) && Path(`/login`)"',
            f"      service: {code}-drive",
            "      entryPoints: [websecure]",
            f"      middlewares: [{code}-oidc-redirect]",
            "      tls:",
            "        certResolver: letsencrypt",
            "      priority: 300",
            "",
            f"    {code}-admin:",
            f'      rule: "Host(`{domain}`) && PathPrefix(`/admin`)"',
            f"      service: {code}-vmis",
            "      entryPoints: [websecure]",
            "      middlewares: [vmis-strip-admin]",
            "      tls:",
            "        certResolver: letsencrypt",
            "      priority: 200",
            "",
            f"    {code}-api:",
            f'      rule: "Host(`{domain}`) && PathPrefix(`/api`)"',
            f"      service: {code}-vmis",
            "      entryPoints: [websecure]",
            "      tls:",
            "        certResolver: letsencrypt",
            "      priority: 200",
            "",
            f"    {code}-drive:",
            f'      rule: "Host(`{domain}`)"',
            f"      service: {code}-drive",
            "      entryPoints: [websecure]",
            "      tls:",
            "        certResolver: letsencrypt",
            "",
            f"    {code}-http:",
            f'      rule: "Host(`{domain}`)"',
            "      entryPoints: [web]",
            "      middlewares: [redirect-https]",
            f"      service: {code}-drive",
            "",
            "  services:",
            f"    {code}-drive:",
            "      loadBalancer:",
            "        servers:",
            f'          - url: "{nc_url}"',
            f"    {code}-vmis:",
            "      loadBalancer:",
            "        servers:",
            f'          - url: "{vmis_backend_url}"',
        ]
    else:
        lines += [
            "  middlewares:",
            f"    {code}-oidc-redirect:",
            "      redirectRegex:",
            '        regex: ".*"',
            f'        replacement: "https://{domain}/apps/user_oidc/login/1"',
            "        permanent: false",
            "",
            "  routers:",
            f"    {code}-login:",
            f'      rule: "Host(`{domain}`) && Path(`/login`)"',
            f"      service: {code}-drive",
            "      entryPoints: [websecure]",
            f"      middlewares: [{code}-oidc-redirect]",
            "      tls:",
            "        certResolver: letsencrypt",
            "      priority: 300",
            "",
            f"    {code}-drive:",
            f'      rule: "Host(`{domain}`)"',
            f"      service: {code}-drive",
            "      entryPoints: [websecure]",
            "      tls:",
            "        certResolver: letsencrypt",
            "",
            f"    {code}-http:",
            f'      rule: "Host(`{domain}`)"',
            "      entryPoints: [web]",
            "      middlewares: [redirect-https]",
            f"      service: {code}-drive",
            "",
            "  services:",
            f"    {code}-drive:",
            "      loadBalancer:",
            "        servers:",
            f'          - url: "{nc_url}"',
        ]

    return "\n".join(lines) + "\n"


def _ensure_traefik_routes(tenant, is_active: bool) -> bool:
    """
    確保租戶 Traefik 路由檔案存在且內容正確。
    使用 SFTP 寫入 /home/porsche/traefik/dynamic/{code}.yml，Traefik 熱重載。
    回傳 True=路由已生效, False=失敗
    """
    import time
    try:
        import paramiko
        code = tenant.code
        yaml_name = f"{code}.yml" if is_active else f"{code}-test.yml"
        file_path = f"{TRAEFIK_DYNAMIC_DIR}/{yaml_name}"
        expected = _generate_tenant_route_yaml(tenant, is_active)

        client = paramiko.SSHClient()
        client.set_missing_host_key_policy(paramiko.AutoAddPolicy())
        client.connect(settings.DOCKER_SSH_HOST, username=settings.DOCKER_SSH_USER, timeout=15)

        sftp = client.open_sftp()
        needs_write = True
        try:
            with sftp.open(file_path, "r") as f:
                existing = f.read().decode()
            if existing == expected:
                needs_write = False
                logger.info(f"Traefik route {yaml_name}: already correct")
        except FileNotFoundError:
            logger.info(f"Traefik route {yaml_name}: not found, creating")

        if needs_write:
            with sftp.open(file_path, "w") as f:
                f.write(expected)
            logger.info(f"Traefik route {yaml_name}: written")

        sftp.close()
        client.close()

        # 驗證 Traefik 已載入路由（最多等待 4 秒）
        route_name = f"{code}-drive@file"
        for attempt in range(2):
            try:
                resp = httpx.get(
                    f"{TRAEFIK_API_URL}/api/http/routers/{route_name}",
                    timeout=5.0,
                )
                if resp.status_code == 200 and resp.json().get("status") == "enabled":
                    logger.info(f"Traefik route {route_name}: enabled ✓")
                    return True
            except Exception:
                pass
            if attempt == 0:
                time.sleep(2)

        logger.warning(f"Traefik route {route_name}: not visible in API after write")
        return False

    except Exception as e:
        logger.error(f"_ensure_traefik_routes {tenant.code}: {e}")
        return False


# ─── Keycloak helpers ────────────────────────────────────────────────────────

def _check_kc_realm(host: str, realm: str) -> Optional[bool]:
    """
    None  = realm 不存在（未設定）
    True  = realm 存在且可連線
    False = 連線失敗
    """
    try:
        resp = httpx.get(
            f"https://{host}/realms/{realm}/.well-known/openid-configuration",
            timeout=5.0,
        )
        if resp.status_code == 200:
            return True
        if resp.status_code == 404:
            return None
        return False
    except Exception as e:
        logger.warning(f"KC realm check failed {host}/{realm}: {e}")
        return False


def _create_kc_realm(realm: str, tenant_name: str):
    from app.services.keycloak_client import get_keycloak_client
    kc = get_keycloak_client()
    kc.create_realm(realm, tenant_name)
    kc.update_realm_token_settings(realm, access_code_lifespan=600)


def _ensure_kc_drive_client(realm: str, domain: str) -> Optional[str]:
    """
    確保 Keycloak realm 中存在 'drive' confidential client（NC OIDC 用）。
    回傳 client_secret，若失敗回傳 None。
    """
    try:
        from app.services.keycloak_client import get_keycloak_client
        kc = get_keycloak_client()
        # redirectUri 為根路徑（NC 直接服務租戶網域，無路徑前綴）
        status = kc.create_confidential_client(
            realm, "drive", [f"https://{domain}/*"]
        )
        if status in ("exists", "created"):
            return kc.get_client_secret(realm, "drive")
        logger.error(f"Failed to ensure drive client in realm {realm}: {status}")
        return None
    except Exception as e:
        logger.error(f"_ensure_kc_drive_client {realm}: {e}")
        return None


def _nc_db_check(container_name: str, pg_host: str, pg_db: str, nc_domain: str, pg_port: int = PG_PORT_TRIAL) -> bool:
    """
    驗證 NC 是否正確安裝並使用 PostgreSQL。
    若偵測到 SQLite（常見於 volumes 持久但重新部署的情況），自動修復：
      1. 刪除 config.php
      2. 清空 PostgreSQL DB schema
      3. 清除 data 目錄中的 SQLite 檔案
      4. 重新執行 occ maintenance:install 使用 pgsql
      5. 設定 overwritehost
    回傳 True=已正確使用 pgsql, False=仍有問題
    """
    try:
        import paramiko, json as _json, psycopg2
        client = paramiko.SSHClient()
        client.set_missing_host_key_policy(paramiko.AutoAddPolicy())
        client.connect(settings.DOCKER_SSH_HOST, username=settings.DOCKER_SSH_USER, timeout=15)

        # 取得 NC 安裝狀態
        _, stdout, _ = client.exec_command(
            f"docker exec -u www-data {container_name} "
            f"php /var/www/html/occ status --output=json 2>/dev/null"
        )
        status_raw = stdout.read().decode().strip()

        installed = False
        try:
            installed = _json.loads(status_raw).get("installed", False)
        except Exception:
            pass

        if not installed:
            logger.info(f"NC {container_name}: not installed yet, installing with pgsql...")
            # 重置 PostgreSQL DB schema（確保乾淨狀態）
            conn = psycopg2.connect(
                host=pg_host, port=pg_port, dbname=pg_db,
                user=PG_USER, password=PG_PASS, connect_timeout=5,
            )
            conn.autocommit = True
            cur = conn.cursor()
            cur.execute("DROP SCHEMA public CASCADE; CREATE SCHEMA public; GRANT ALL ON SCHEMA public TO PUBLIC;")
            conn.close()
            logger.info(f"PostgreSQL {pg_db}@{pg_host}: schema reset for fresh install")

            install_cmd = (
                f"docker exec -u www-data {container_name} php /var/www/html/occ maintenance:install "
                f"-n --admin-user admin --admin-pass NC1qaz2wsx "
                f"--database pgsql --database-name '{pg_db}' "
                f"--database-user {PG_USER} --database-pass {PG_PASS} "
                f"--database-host '{pg_host}:{pg_port}' 2>&1"
            )
            _, stdout_inst, _ = client.exec_command(install_cmd)
            stdout_inst.channel.settimeout(120)
            try:
                install_out = stdout_inst.read().decode().strip()
            except Exception:
                install_out = ""
            logger.info(f"NC fresh install output: {install_out}")

            # 設定 overwritehost
            for cfg_cmd in [
                f"docker exec -u www-data {container_name} php /var/www/html/occ config:system:set overwritehost --value={nc_domain}",
                f"docker exec -u www-data {container_name} php /var/www/html/occ config:system:set overwrite.cli.url --value=https://{nc_domain}",
            ]:
                _, out_cfg, _ = client.exec_command(cfg_cmd)
                out_cfg.channel.settimeout(30)
                try:
                    out_cfg.read()
                except Exception:
                    pass

            client.close()
            success = "successfully installed" in install_out
            if success:
                logger.info(f"NC {container_name}: installed with pgsql ✓")
            else:
                logger.error(f"NC {container_name}: fresh install failed: {install_out}")
            return success

        # 檢查 dbtype
        _, stdout2, _ = client.exec_command(
            f"docker exec -u www-data {container_name} "
            f"php /var/www/html/occ config:system:get dbtype 2>/dev/null"
        )
        dbtype = stdout2.read().decode().strip()
        client.close()

        if dbtype == "pgsql":
            logger.info(f"NC {container_name}: already using pgsql ✓")
            return True

        # ─── 偵測到 SQLite，自動修復 ───────────────────────────────────────
        logger.warning(f"NC {container_name}: dbtype={dbtype}, fixing to pgsql...")

        # 1. 刪除 config.php 和 SQLite 殘留資料
        client2 = paramiko.SSHClient()
        client2.set_missing_host_key_policy(paramiko.AutoAddPolicy())
        client2.connect(settings.DOCKER_SSH_HOST, username=settings.DOCKER_SSH_USER, timeout=15)
        for cmd in [
            f"docker exec {container_name} rm -f /var/www/html/config/config.php",
            f"docker exec {container_name} sh -c "
            f"'rm -rf /var/www/html/data/admin /var/www/html/data/appdata_* "
            f"/var/www/html/data/*.db /var/www/html/data/nextcloud.log'",
        ]:
            _, out, _ = client2.exec_command(cmd)
            out.channel.recv_exit_status()
            logger.info(f"Cleanup: {cmd}")

        # 2. 重置 PostgreSQL DB schema
        conn = psycopg2.connect(
            host=pg_host, port=pg_port, dbname=pg_db,
            user=PG_USER, password=PG_PASS, connect_timeout=5,
        )
        conn.autocommit = True
        cur = conn.cursor()
        cur.execute("DROP SCHEMA public CASCADE; CREATE SCHEMA public; GRANT ALL ON SCHEMA public TO PUBLIC;")
        conn.close()
        logger.info(f"PostgreSQL {pg_db}@{pg_host}: schema reset")

        # 3. 重新安裝 NC 使用 PostgreSQL
        install_cmd = (
            f"docker exec -u www-data {container_name} php /var/www/html/occ maintenance:install "
            f"-n --admin-user admin --admin-pass NC1qaz2wsx "
            f"--database pgsql --database-name '{pg_db}' "
            f"--database-user {PG_USER} --database-pass {PG_PASS} "
            f"--database-host '{pg_host}:{pg_port}' 2>&1"
        )
        _, stdout3, _ = client2.exec_command(install_cmd)
        stdout3.channel.recv_exit_status()
        install_out = stdout3.read().decode().strip()
        logger.info(f"NC reinstall output: {install_out}")

        # 4. 設定 overwritehost
        for cfg_cmd in [
            f"docker exec -u www-data {container_name} php /var/www/html/occ config:system:set overwritehost --value={nc_domain}",
            f"docker exec -u www-data {container_name} php /var/www/html/occ config:system:set overwrite.cli.url --value=https://{nc_domain}",
        ]:
            _, out4, _ = client2.exec_command(cfg_cmd)
            out4.channel.recv_exit_status()

        client2.close()
        success = "successfully installed" in install_out
        if success:
            logger.info(f"NC {container_name}: fixed to pgsql ✓")
        else:
            logger.error(f"NC {container_name}: reinstall failed: {install_out}")
        return success

    except Exception as e:
        logger.error(f"_nc_db_check {container_name}: {e}")
        return False


def _nc_initialized(container_name: str) -> bool:
    """
    以 force_language == zh_TW 作為 NC 已完成初始化的判斷依據。
    """
    try:
        import paramiko
        client = paramiko.SSHClient()
        client.set_missing_host_key_policy(paramiko.AutoAddPolicy())
        client.connect(settings.DOCKER_SSH_HOST, username=settings.DOCKER_SSH_USER, timeout=15)
        _, stdout, _ = client.exec_command(
            f"docker exec -u www-data {container_name} "
            f"php /var/www/html/occ config:system:get force_language 2>/dev/null"
        )
        result = stdout.read().decode().strip()
        client.close()
        return result == "zh_TW"
    except Exception as e:
        logger.warning(f"NC initialized check failed {container_name}: {e}")
        return False


def _nc_initialize(
    container_name: str, kc_host: str, realm: str, client_secret: str, nc_domain: str,
    oo_container: str, oo_url: str, is_active: bool = False,
) -> bool:
    """
    NC 容器首次完整初始化：
    Init-1: 語言設定（zh_TW）
    Init-2: 安裝必要 Apps（contacts / calendar / mail / onlyoffice）
    Init-3: OIDC Provider 設定（呼叫 _configure_nc_oidc）
    Init-4: SSO 強制模式（allow_multiple_user_backends=0）
    Init-5: OnlyOffice 整合設定
    ⚠️ Init-4 必須在 Init-3 之後執行，否則無法登入
    """
    try:
        import paramiko
        client = paramiko.SSHClient()
        client.set_missing_host_key_policy(paramiko.AutoAddPolicy())
        client.connect(settings.DOCKER_SSH_HOST, username=settings.DOCKER_SSH_USER, timeout=15)

        redis_host = REDIS_HOST_ACTIVE if is_active else REDIS_HOST_TRIAL

        # Init-1: 語言設定
        for cfg_key, cfg_val in [
            ("default_language", "zh_TW"),
            ("default_locale", "zh_TW"),
            ("force_language", "zh_TW"),
            ("force_locale", "zh_TW"),
        ]:
            _, out, _ = client.exec_command(
                f"docker exec -u www-data {container_name} "
                f"php /var/www/html/occ config:system:set {cfg_key} --value={cfg_val} 2>&1"
            )
            out.channel.recv_exit_status()

        # Init-1b: Redis + APCu memcache（OIDC session 持久化必須）
        for cfg_cmd in [
            f"docker exec -u www-data {container_name} php /var/www/html/occ config:system:set memcache.local --value='\\OC\\Memcache\\APCu' 2>&1",
            f"docker exec -u www-data {container_name} php /var/www/html/occ config:system:set memcache.distributed --value='\\OC\\Memcache\\Redis' 2>&1",
            f"docker exec -u www-data {container_name} php /var/www/html/occ config:system:set memcache.locking --value='\\OC\\Memcache\\Redis' 2>&1",
            f"docker exec -u www-data {container_name} php /var/www/html/occ config:system:set redis host --value={redis_host} 2>&1",
            f"docker exec -u www-data {container_name} php /var/www/html/occ config:system:set redis port --value={REDIS_PORT} --type=integer 2>&1",
            f"docker exec -u www-data {container_name} php /var/www/html/occ config:system:set redis password --value={REDIS_PASS} 2>&1",
        ]:
            _, out, _ = client.exec_command(cfg_cmd)
            out.channel.recv_exit_status()
        logger.info(f"NC {container_name}: Redis memcache configured (host={redis_host})")

        # Init-2: 安裝必要 Apps（已安裝時不影響結果）
        for app in ["contacts", "calendar", "mail", "onlyoffice"]:
            _, out, _ = client.exec_command(
                f"docker exec -u www-data {container_name} "
                f"php /var/www/html/occ app:install {app} 2>&1"
            )
            out.channel.recv_exit_status()
            text = out.read().decode().strip()
            logger.info(f"NC {container_name} app:install {app}: {text}")

        client.close()

        # Init-3: OIDC Provider 設定（複用現有函式）
        _configure_nc_oidc(container_name, kc_host, realm, client_secret, nc_domain)

        # Init-4: 強制 SSO（禁用本地登入）+ Init-5: OnlyOffice 整合設定
        client2 = paramiko.SSHClient()
        client2.set_missing_host_key_policy(paramiko.AutoAddPolicy())
        client2.connect(settings.DOCKER_SSH_HOST, username=settings.DOCKER_SSH_USER, timeout=15)

        _, out, _ = client2.exec_command(
            f"docker exec -u www-data {container_name} "
            f"php /var/www/html/occ config:app:set user_oidc allow_multiple_user_backends --value=0 2>&1"
        )
        out.channel.recv_exit_status()

        # Init-5: OnlyOffice 整合設定
        from app.core.config import settings as _cfg
        for oo_key, oo_val in [
            ("DocumentServerUrl",      f"{oo_url}/"),
            ("DocumentServerInternalUrl", f"http://{oo_container}/"),
            ("StorageUrl",             f"https://{nc_domain}/"),
            ("jwt_secret",             _cfg.OO_JWT_SECRET),
        ]:
            _, out, _ = client2.exec_command(
                f"docker exec -u www-data {container_name} "
                f"php /var/www/html/occ config:app:set onlyoffice {oo_key} --value='{oo_val}' 2>&1"
            )
            out.channel.recv_exit_status()

        client2.close()

        logger.info(f"NC {container_name}: initialization complete ✓")
        return True
    except Exception as e:
        logger.error(f"NC initialization failed {container_name}: {e}")
        return False


def _nc_oidc_configured(container_name: str) -> bool:
    """檢查 NC 容器是否已設定 OIDC provider（user_oidc app 已安裝且有 provider）"""
    try:
        import paramiko
        client = paramiko.SSHClient()
        client.set_missing_host_key_policy(paramiko.AutoAddPolicy())
        client.connect(
            settings.DOCKER_SSH_HOST,
            username=settings.DOCKER_SSH_USER,
            timeout=15,
        )
        _, stdout, _ = client.exec_command(
            f"docker exec -u www-data {container_name} "
            f"php /var/www/html/occ user_oidc:providers 2>/dev/null | grep -q clientId && echo yes || echo no"
        )
        result = stdout.read().decode().strip()
        client.close()
        return result == "yes"
    except Exception as e:
        logger.warning(f"NC OIDC check failed for {container_name}: {e}")
        return False


def _configure_nc_oidc(
    container_name: str, kc_host: str, realm: str, client_secret: str, nc_domain: str
) -> bool:
    """
    設定 NC OIDC provider（使用 provider name='drive'，對應 Keycloak client_id=drive）。
    同時設定 overwritehost 確保 OIDC callback URL 正確。
    """
    try:
        import paramiko
        client = paramiko.SSHClient()
        client.set_missing_host_key_policy(paramiko.AutoAddPolicy())
        client.connect(
            settings.DOCKER_SSH_HOST,
            username=settings.DOCKER_SSH_USER,
            timeout=15,
        )
        discovery = f"https://{kc_host}/realms/{realm}/.well-known/openid-configuration"

        # 設定 overwritehost（OIDC callback 必須）
        for cfg_cmd in [
            f"docker exec -u www-data {container_name} php /var/www/html/occ config:system:set overwritehost --value={nc_domain}",
            f"docker exec -u www-data {container_name} php /var/www/html/occ config:system:set overwrite.cli.url --value=https://{nc_domain}",
        ]:
            _, out, _ = client.exec_command(cfg_cmd)
            out.channel.recv_exit_status()

        # 啟用 user_oidc app
        _, out2, _ = client.exec_command(
            f"docker exec -u www-data {container_name} php /var/www/html/occ app:enable user_oidc 2>&1"
        )
        out2.channel.recv_exit_status()

        # 設定 OIDC provider（name=drive，與 Keycloak client_id 一致）
        oidc_cmd = (
            f"docker exec -u www-data {container_name} php /var/www/html/occ user_oidc:provider drive "
            f"--clientid=drive "
            f"--clientsecret={client_secret} "
            f"--discoveryuri={discovery} "
            f"--mapping-uid=preferred_username "
            f"--mapping-display-name=name "
            f"--mapping-email=email "
            f"--unique-uid=0 "
            f"--check-bearer=0 --send-id-token-hint=1 2>&1"
        )
        _, stdout, _ = client.exec_command(oidc_cmd)
        stdout.channel.recv_exit_status()
        out_text = stdout.read().decode().strip()
        client.close()
        logger.info(f"NC OIDC configure output: {out_text}")
        return True
    except Exception as e:
        logger.error(f"NC OIDC configure failed for {container_name}: {e}")
        return False


# ─── PostgreSQL helpers ───────────────────────────────────────────────────────

def _ensure_nc_db(host: str, dbname: str, port: int = PG_PORT_TRIAL) -> bool:
    """
    確保 NC 用的 PostgreSQL DB 存在，並授予 public schema 建表權限給所有用戶。
    NC Docker 會自動建立 oc_admin* 用戶，需預先開放 public schema CREATE 權限。
    """
    try:
        import psycopg2
        # Connect to postgres DB to create NC DB
        conn = psycopg2.connect(
            host=host, port=port, dbname="postgres",
            user=PG_USER, password=PG_PASS,
            connect_timeout=5,
        )
        conn.autocommit = True
        cur = conn.cursor()
        cur.execute("SELECT 1 FROM pg_database WHERE datname = %s", (dbname,))
        if not cur.fetchone():
            cur.execute(f'CREATE DATABASE "{dbname}" OWNER {PG_USER}')
            logger.info(f"Created database {dbname}@{host}")
        conn.close()

        # Grant CREATE on public schema to all users (for NC oc_admin* users)
        conn2 = psycopg2.connect(
            host=host, port=port, dbname=dbname,
            user=PG_USER, password=PG_PASS,
            connect_timeout=5,
        )
        conn2.autocommit = True
        cur2 = conn2.cursor()
        cur2.execute("GRANT ALL ON SCHEMA public TO PUBLIC")
        conn2.close()
        return True
    except Exception as e:
        logger.error(f"_ensure_nc_db {dbname}@{host}: {e}")
        return False


def _get_pg_db_size_gb(host: str, dbname: str, port: int = PG_PORT_TRIAL) -> Optional[float]:
    try:
        import psycopg2
        conn = psycopg2.connect(
            host=host, port=port, dbname=dbname,
            user=PG_USER, password=PG_PASS,
            connect_timeout=5,
        )
        cur = conn.cursor()
        cur.execute("SELECT pg_database_size(%s)", (dbname,))
        size_bytes = cur.fetchone()[0]
        cur.close()
        conn.close()
        return round(size_bytes / (1024 ** 3), 3)
    except Exception as e:
        logger.warning(f"PG size check failed {dbname}@{host}: {e}")
        return None


# ─── Email notification ───────────────────────────────────────────────────────

def _get_admin_emails(db: Session) -> list[str]:
    """取得所有管理員租戶下所有啟用帳號的 notification_email"""
    from app.models.account import Account
    rows = (
        db.query(Account.email)
        .join(Tenant, Account.tenant_id == Tenant.id)
        .filter(
            Tenant.is_manager == True,
            Account.is_active == True,
            Account.email != None,
            Account.email != "",
        )
        .all()
    )
    return list({r.email for r in rows})


def _send_failure_alert(
    tenant_code: str,
    tenant_name: str,
    domain: str,
    failed_items: list[str],
    admin_emails: list[str],
):
    """任何檢查項目 False 時，統一發送告警 email 給所有管理員"""
    if not admin_emails:
        logger.warning(f"No admin emails for failure alert on {tenant_code}")
        return
    try:
        item_lines = "\n".join(f"  ✗ {item}" for item in failed_items)
        body = (
            f"【Virtual MIS 告警】租戶服務異常\n\n"
            f"租戶代碼 : {tenant_code}\n"
            f"租戶名稱 : {tenant_name}\n"
            f"網域     : {domain}\n"
            f"時間     : {now_tw().strftime('%Y-%m-%d %H:%M:%S')}\n\n"
            f"異常項目:\n{item_lines}\n\n"
            f"請盡速登入 Virtual MIS 後台確認並處理。"
        )
        msg = MIMEText(body, "plain", "utf-8")
        msg["Subject"] = f"[VirtualMIS] 服務異常告警 — {tenant_code} ({domain})"
        msg["From"] = f"vmis-alert@{settings.MAIL_MX_HOST}"
        msg["To"] = ", ".join(admin_emails)

        with smtplib.SMTP(settings.MAIL_MX_HOST, 25, timeout=10) as smtp:
            smtp.sendmail(msg["From"], admin_emails, msg.as_string())
        logger.info(f"Failure alert sent to {admin_emails} for {tenant_code}: {failed_items}")
    except Exception as e:
        logger.error(f"Failed to send failure alert for {tenant_code}: {e}")


# ─── Main check ──────────────────────────────────────────────────────────────

def run_tenant_check(schedule_log_id: int, db: Session):
    from app.services.mail_client import MailClient
    from app.services.docker_client import DockerClient

    tenants = db.query(Tenant).filter(Tenant.is_active == True).all()
    mail = MailClient()
    docker = DockerClient()
    admin_emails = _get_admin_emails(db)

    for tenant in tenants:
        realm = tenant.code
        is_active = tenant.status == "active"

        deploy_name = tenant.code if is_active else f"{tenant.code}-test"
        nc_name = f"nc-{tenant.code}" if is_active else f"nc-{tenant.code}-test"
        oo_name = f"oo-{tenant.code}" if is_active else f"oo-{tenant.code}-test"
        kc_host = KC_HOST_ACTIVE if is_active else KC_HOST_TRIAL
        pg_host = PG_HOST_ACTIVE if is_active else PG_HOST_TRIAL
        pg_port = PG_PORT_ACTIVE if is_active else PG_PORT_TRIAL

        result = TenantScheduleResult(
            schedule_log_id=schedule_log_id,
            tenant_id=tenant.id,
            recorded_at=now_tw(),
        )
        fail_reasons = []

        # ── [1] Traefik 路由檔案確認 ─────────────────────────────────────────
        # 使用 file provider，scheduler 直接寫入 /home/porsche/traefik/dynamic/{code}.yml
        try:
            ok = _ensure_traefik_routes(tenant, is_active)
            result.traefik_status = ok
            if not ok:
                fail_reasons.append("traefik: route not loaded after write")
            result.traefik_done_at = now_tw()
        except Exception as e:
            result.traefik_status = False
            result.traefik_done_at = now_tw()
            fail_reasons.append(f"traefik: {e}")


        # ── [2] SSO (Keycloak realm + drive client) ──────────────────────────
        kc_drive_secret: Optional[str] = None
        try:
            sso_state = _check_kc_realm(kc_host, realm)
            if sso_state is None:
                # Realm 不存在 → 建立 realm + drive client
                try:
                    _create_kc_realm(realm, tenant.name)
                    kc_drive_secret = _ensure_kc_drive_client(realm, tenant.domain)
                    result.sso_result = True if kc_drive_secret else False
                    if not kc_drive_secret:
                        fail_reasons.append("sso: realm created but drive client failed")
                except Exception as ce:
                    result.sso_result = False
                    fail_reasons.append(f"sso create: {ce}")
            elif sso_state is True:
                # Realm 存在 → 確保 drive client 存在，並確保 token 逾時設定
                kc_drive_secret = _ensure_kc_drive_client(realm, tenant.domain)
                result.sso_result = True
                if not kc_drive_secret:
                    fail_reasons.append("sso: drive client missing/failed")
                    result.sso_result = False
                else:
                    try:
                        from app.services.keycloak_client import get_keycloak_client
                        get_keycloak_client().update_realm_token_settings(realm, access_code_lifespan=600)
                    except Exception:
                        pass
            else:
                result.sso_result = False
                fail_reasons.append("sso: realm unreachable")
            result.sso_done_at = now_tw()
        except Exception as e:
            result.sso_result = False
            result.sso_done_at = now_tw()
            fail_reasons.append(f"sso: {e}")

        # ── [3] Mailbox domain ───────────────────────────────────────────────
        try:
            if mail.domain_exists(tenant.domain):
                result.mailbox_result = True
            else:
                # Domain 未設定 → 建立
                ok = mail.create_domain(tenant.domain)
                result.mailbox_result = True if ok else False
                if not ok:
                    fail_reasons.append("mailbox: create domain failed")
            result.mailbox_done_at = now_tw()
        except Exception as e:
            result.mailbox_result = False
            result.mailbox_done_at = now_tw()
            fail_reasons.append(f"mailbox: {e}")

        # ── [4] NC container + DB 驗證 + OIDC 設定 ─────────────────────────
        pg_db = f"nc_{tenant.code}_db"
        try:
            nc_state = docker.check_container_ssh(nc_name)
            if nc_state is None:
                # 容器不存在 → 確保 docker-compose.yml + DB + 部署
                logger.info(f"NC {nc_name}: not found, ensuring compose/DB and deploying")
                _ensure_tenant_compose(tenant, is_active)
                _ensure_nc_db(pg_host, pg_db, pg_port)
                ok = docker.ssh_compose_up(deploy_name)
                result.nc_result = True if ok else False
                if not ok:
                    fail_reasons.append("nc: deploy failed")
                else:
                    if not _nc_db_check(nc_name, pg_host, pg_db, tenant.domain, pg_port):
                        result.nc_result = False
                        fail_reasons.append("nc: installed but not using pgsql")
            elif nc_state is False:
                # 容器存在但已停止 → 重啟
                logger.info(f"NC {nc_name}: stopped, restarting")
                ok = docker.ssh_compose_up(deploy_name)
                result.nc_result = True if ok else False
                if not ok:
                    fail_reasons.append("nc: start failed")
            else:
                # 容器正常運行 → 驗證 DB 類型
                db_ok = _nc_db_check(nc_name, pg_host, pg_db, tenant.domain, pg_port)
                if not db_ok:
                    result.nc_result = False
                    fail_reasons.append("nc: DB check failed (possible sqlite3 issue)")
                else:
                    result.nc_result = True
                    if kc_drive_secret:
                        if not _nc_initialized(nc_name):
                            oo_url = (f"https://office-{tenant.code}.ease.taipei" if is_active
                                      else f"https://office-{tenant.code}.lab.taipei")
                            ok = _nc_initialize(nc_name, kc_host, realm, kc_drive_secret, tenant.domain,
                                                oo_name, oo_url, is_active)
                            if not ok:
                                fail_reasons.append("nc: initialization failed")
                        else:
                            ok = _configure_nc_oidc(nc_name, kc_host, realm, kc_drive_secret, tenant.domain)
                            if not ok:
                                fail_reasons.append("nc: OIDC sync failed")
            result.nc_done_at = now_tw()
        except Exception as e:
            result.nc_result = False
            result.nc_done_at = now_tw()
            fail_reasons.append(f"nc: {e}")

        # ── [5] OO container ─────────────────────────────────────────────────
        try:
            oo_state = docker.check_container_ssh(oo_name)
            if oo_state is None:
                ok = docker.ssh_compose_up(deploy_name)
                result.office_result = True if ok else False
                if not ok:
                    fail_reasons.append("oo: deploy failed")
            elif oo_state is False:
                ok = docker.ssh_compose_up(deploy_name)
                result.office_result = True if ok else False
                if not ok:
                    fail_reasons.append("oo: start failed")
            else:
                result.office_result = True
            result.office_done_at = now_tw()
        except Exception as e:
            result.office_result = False
            result.office_done_at = now_tw()
            fail_reasons.append(f"oo: {e}")

        # ── [6] Quota (OO disk + PG DB size) ────────────────────────────────
        try:
            oo_gb = docker.get_oo_disk_usage_gb(oo_name) or 0.0
            pg_gb = _get_pg_db_size_gb(pg_host, pg_db, pg_port) or 0.0
            result.quota_usage = round(oo_gb + pg_gb, 3)
        except Exception as e:
            logger.warning(f"Quota check failed for {tenant.code}: {e}")

        if fail_reasons:
            result.fail_reason = "; ".join(fail_reasons)

        # 任何項目 False → 統一發送告警給所有管理員
        failed_items = []
        if result.traefik_status is False:
            failed_items.append("Traefik 路由")
        if result.sso_result is False:
            failed_items.append("SSO (Keycloak Realm)")
        if result.mailbox_result is False:
            failed_items.append("Mailbox Domain")
        if result.nc_result is False:
            failed_items.append("Nextcloud 容器")
        if result.office_result is False:
            failed_items.append("OnlyOffice 容器")
        if failed_items:
            _send_failure_alert(
                tenant.code, tenant.name, tenant.domain, failed_items, admin_emails
            )

        db.add(result)

    db.commit()
    logger.info(f"Tenant check done: {len(tenants)} tenants processed")