vps-manager/app/services/action_service.py

"""云资源管理动作服务

把「资产 → 账号 → 平台适配器」解析成一次真实的云端写操作。

与只读同步的关键区别：这里的每个动作都会改变云端真实资源——关机会中断线上服务、
续费会真实扣款。因此三条硬约束：
1. 路由层强制 require_api_key（外部流量无 key 一律 401）
2. 每次动作都写 EventLog（source="action"），成功与失败都记，且用独立 Session 提交，
   避免主流程回滚时把审计记录一起丢掉
3. 适配器不支持该动作时如实抛 400，绝不静默 no-op 让调用方以为成功了
"""

import logging

from fastapi import HTTPException, status
from sqlmodel import Session, select

from app.adapters import registry
from app.adapters.base import BaseAdapter, cloud_error_text
from app.core.timeutils import utcnow
from app.models.asset import Asset, AssetType, DomainDetail
from app.services.sync_service import _build_adapter_for_account, _missing_config

logger = logging.getLogger("vps-manager.action")

#: 动作名 -> 适配器方法名
INSTANCE_ACTIONS = {
    "start": "start_instance",
    "stop": "stop_instance",
    "reboot": "reboot_instance",
    "renew": "renew_instance",
}

#: 各动作的中文标签，用于审计日志与前端提示
ACTION_LABELS = {
    "start": "开机",
    "stop": "关机",
    "reboot": "重启",
    "renew": "续费",
}


def audit(asset_id: int | None, message: str, level: str = "info") -> None:
    """写审计日志（独立 Session，主流程回滚也不丢）

    审计失败绝不能让业务动作失败——云端指令可能已经下发成功，
    此时抛错会让调用方误判为「没执行」，比丢一条日志危险得多。
    """
    from app.database import metrics_engine
    from app.models.monitor import EventLog

    try:
        with Session(metrics_engine) as s:
            s.add(EventLog(level=level, source="action", asset_id=asset_id, message=message))
            s.commit()
    except Exception as e:  # noqa: BLE001
        logger.warning("审计日志写入失败 asset_id=%s err=%s", asset_id, e)


def _resolve(session: Session, asset_id: int) -> tuple[Asset, BaseAdapter, str]:
    """解析资产对应的适配器，返回 (asset, adapter, provider_label)"""
    asset = session.get(Asset, asset_id)
    if not asset:
        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="资产不存在")
    if not asset.account_id:
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST,
            detail="该资产未关联账号，无法确定使用哪套云凭证执行动作",
        )
    from app.models.asset import Account

    account = session.get(Account, asset.account_id)
    if not account:
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST, detail="资产关联的账号已不存在"
        )
    provider, adapter = _build_adapter_for_account(session, account)
    missing = _missing_config(adapter)
    if missing:
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST,
            detail=f"缺少凭证配置：{', '.join(missing)}（请先在账号编辑里填写 API 配置）",
        )
    return asset, adapter, provider.slug


def _require_capability(adapter: BaseAdapter, method_name: str, label: str) -> None:
    if not adapter.capabilities().get(method_name):
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST,
            detail=f"{type(adapter).__name__} 不支持{label}",
        )


def perform_instance_action(
    session: Session, asset_id: int, action: str, period_months: int = 1
) -> dict:
    """对 VPS 资产执行 开机/关机/重启/续费"""
    if action not in INSTANCE_ACTIONS:
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST,
            detail=f"未知动作：{action}（可选 {'/'.join(INSTANCE_ACTIONS)}）",
        )
    asset, adapter, provider_slug = _resolve(session, asset_id)
    if asset.asset_type != AssetType.VPS:
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST,
            detail=f"仅 VPS 资产支持该动作（当前类型：{asset.asset_type.value}）",
        )
    if not asset.external_id:
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST,
            detail="该资产没有云端实例 ID（external_id），可能是手工登记的，无法下发指令",
        )

    method_name = INSTANCE_ACTIONS[action]
    label = ACTION_LABELS[action]
    _require_capability(adapter, method_name, label)

    target = f"{provider_slug}/{asset.name}({asset.external_id})"
    audit(asset_id, f"{label}动作下发：{target}" + (f" 周期={period_months}月" if action == "renew" else ""))
    try:
        method = getattr(adapter, method_name)
        result = method(asset.external_id, period_months) if action == "renew" else method(asset.external_id)
    except NotImplementedError:
        audit(asset_id, f"{label}失败：适配器未实现 {target}", level="warning")
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST, detail=f"{provider_slug} 暂不支持{label}"
        ) from None
    except Exception as e:  # noqa: BLE001
        msg = cloud_error_text(e)
        audit(asset_id, f"{label}异常：{target} err={msg}", level="error")
        raise HTTPException(
            status_code=status.HTTP_502_BAD_GATEWAY, detail=f"{label}失败：{msg}"
        ) from e

    ok = bool(result.get("ok"))
    audit(
        asset_id,
        f"{label}{'成功' if ok else '被拒'}：{target} msg={result.get('message')}",
        level="info" if ok else "error",
    )
    return {
        "ok": ok,
        "action": action,
        "asset_id": asset_id,
        "provider": provider_slug,
        "external_id": asset.external_id,
        "message": result.get("message", ""),
        "at": utcnow().isoformat(),
    }


def _resolve_zone(session: Session, asset_id: int) -> tuple[Asset, BaseAdapter, str, str]:
    """解析域名资产的 DNS zone 标识，返回 (asset, adapter, provider_slug, zone_id)

    zone_id 取 Asset.external_id：Cloudflare 是 zone uuid，阿里云/腾讯云/Vultr/DO 是域名字符串，
    两者的 list_dns_records 都已按各自约定实现。
    """
    asset, adapter, provider_slug = _resolve(session, asset_id)
    if asset.asset_type != AssetType.DOMAIN:
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST,
            detail=f"仅域名资产支持 DNS 记录管理（当前类型：{asset.asset_type.value}）",
        )
    zone_id = asset.external_id
    if not zone_id:
        detail = session.exec(
            select(DomainDetail).where(DomainDetail.asset_id == asset_id)
        ).first()
        zone_id = (detail.domain_name if detail else None) or asset.name
    if not zone_id:
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST,
            detail="无法确定该域名的 zone 标识（external_id 与 domain_name 均为空）",
        )
    return asset, adapter, provider_slug, zone_id


def list_dns_records(session: Session, asset_id: int) -> dict:
    asset, adapter, provider_slug, zone_id = _resolve_zone(session, asset_id)
    _require_capability(adapter, "list_dns_records", "DNS 记录查询")
    try:
        records = [r.to_dict() for r in adapter.list_dns_records(zone_id)]
    except NotImplementedError:
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST, detail=f"{provider_slug} 暂不支持 DNS 记录查询"
        ) from None
    except Exception as e:  # noqa: BLE001
        raise HTTPException(
            status_code=status.HTTP_502_BAD_GATEWAY, detail=f"查询 DNS 记录失败：{cloud_error_text(e)}"
        ) from e
    return {"asset_id": asset_id, "provider": provider_slug, "zone_id": zone_id, "records": records}


def _dns_write(
    session: Session,
    asset_id: int,
    method_name: str,
    label: str,
    args: tuple,
) -> dict:
    asset, adapter, provider_slug, zone_id = _resolve_zone(session, asset_id)
    _require_capability(adapter, method_name, label)
    target = f"{provider_slug}/{zone_id}"
    audit(asset_id, f"DNS {label}下发：{target} args={args[1:]}")
    try:
        result = getattr(adapter, method_name)(*args)
    except NotImplementedError:
        audit(asset_id, f"DNS {label}失败：适配器未实现 {target}", level="warning")
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST, detail=f"{provider_slug} 暂不支持{label}"
        ) from None
    except Exception as e:  # noqa: BLE001
        msg = cloud_error_text(e)
        audit(asset_id, f"DNS {label}异常：{target} err={msg}", level="error")
        raise HTTPException(
            status_code=status.HTTP_502_BAD_GATEWAY, detail=f"{label}失败：{msg}"
        ) from e
    ok = bool(result.get("ok"))
    audit(
        asset_id,
        f"DNS {label}{'成功' if ok else '被拒'}：{target} msg={result.get('message')}",
        level="info" if ok else "error",
    )
    return {
        "ok": ok,
        "asset_id": asset_id,
        "provider": provider_slug,
        "zone_id": zone_id,
        "record_id": result.get("record_id"),
        "message": result.get("message", ""),
    }


def create_dns_record(session: Session, asset_id: int, record: dict) -> dict:
    _, _, _, zone_id = _resolve_zone(session, asset_id)
    return _dns_write(session, asset_id, "create_dns_record", "记录创建", (zone_id, record))


def update_dns_record(session: Session, asset_id: int, record_id: str, record: dict) -> dict:
    _, _, _, zone_id = _resolve_zone(session, asset_id)
    return _dns_write(
        session, asset_id, "update_dns_record", "记录更新", (zone_id, record_id, record)
    )


def delete_dns_record(session: Session, asset_id: int, record_id: str) -> dict:
    _, _, _, zone_id = _resolve_zone(session, asset_id)
    return _dns_write(
        session, asset_id, "delete_dns_record", "记录删除", (zone_id, record_id)
    )


def asset_actions(session: Session, asset_id: int) -> dict:
    """返回某资产当前可执行的动作（前端据此渲染/禁用按钮）

    手工登记、无 external_id、或适配器不支持的，一律标 False，
    避免用户点了才收到 400。
    """
    try:
        asset, adapter, provider_slug = _resolve(session, asset_id)
    except HTTPException as e:
        return {"asset_id": asset_id, "available": False, "reason": e.detail, "actions": {}}

    caps = adapter.capabilities()
    is_vps = asset.asset_type == AssetType.VPS
    is_domain = asset.asset_type == AssetType.DOMAIN
    has_external = bool(asset.external_id)
    actions = {
        name: bool(is_vps and has_external and caps.get(method))
        for name, method in INSTANCE_ACTIONS.items()
    }
    actions["dns_records"] = bool(is_domain and caps.get("list_dns_records"))
    actions["dns_write"] = bool(is_domain and caps.get("create_dns_record"))
    return {
        "asset_id": asset_id,
        "provider": provider_slug,
        "sdk_type": getattr(type(adapter), "sdk_types", []),
        "available": any(actions.values()),
        "actions": actions,
    }